0sec-cli 0.18.0 → 0.22.0

package/chunks/{chunk-UR4ELX2Y.js → chunk-2GV4HQKE.js}

@@ -1,19 +1,18 @@
 #!/usr/bin/env node
-import { createRequire as __0secCreateRequire } from "node:module";
-const require = __0secCreateRequire(import.meta.url);
+import { createRequire as __0CreateRequire } from "node:module";
+const require = __0CreateRequire(import.meta.url);
 import {
-  VERSION,
-  homeStateDir
-} from "./chunk-6L26OBWI.js";
+  VERSION
+} from "./chunk-76KSBWOQ.js";
 
 // packages/core/dist/agent/feature-presets.js
 var FP_MOAT_FLAGS = [
-  "0SEC_FEATURE_REACHABILITY_GATE",
-  "0SEC_FEATURE_MULTIMODAL",
-  "0SEC_FEATURE_PUBLISHABILITY_GATE",
-  "0SEC_FEATURE_POV_GATE",
-  "0SEC_FEATURE_POC_GEN_STATIC",
-  "0SEC_FEATURE_CONSENSUS_VERIFY"
+  "ZERO_FEATURE_REACHABILITY_GATE",
+  "ZERO_FEATURE_MULTIMODAL",
+  "ZERO_FEATURE_PUBLISHABILITY_GATE",
+  "ZERO_FEATURE_POV_GATE",
+  "ZERO_FEATURE_POC_GEN_STATIC",
+  "ZERO_FEATURE_CONSENSUS_VERIFY"
 ];
 var FEATURE_PRESETS = Object.freeze({
   "fp-moat": FP_MOAT_FLAGS
@@ -41,7 +40,7 @@ function applyFeaturePreset(preset, env2 = process.env) {
   return { preset, applied, preserved };
 }
 function applyFeaturePresetFromEnv(env2 = process.env) {
-  const raw = env2["0SEC_FEATURE_PRESET"];
+  const raw = env2["ZERO_FEATURE_PRESET"];
   if (!raw)
     return void 0;
   const preset = resolveFeaturePreset(raw);
@@ -63,73 +62,73 @@ var features = {
    *
    * Turn count is also a cost PROXY, and cost is already bounded directly by
    * the token budget, so this never was the control that kept spend in check.
-   * Enable with `0SEC_FEATURE_EARLY_STOP=1` for benchmark or A/B runs where a
+   * Enable with `ZERO_FEATURE_EARLY_STOP=1` for benchmark or A/B runs where a
    * fixed turn budget per attempt is the point.
    */
   get earlyStopRetry() {
-    return env("0SEC_FEATURE_EARLY_STOP", false);
+    return env("ZERO_FEATURE_EARLY_STOP", false);
   },
   /** Detect A-A-A and A-B-A-B loop patterns, inject warning */
   get loopDetection() {
-    return env("0SEC_FEATURE_LOOP_DETECTION", true);
+    return env("ZERO_FEATURE_LOOP_DETECTION", true);
   },
   /** Compress middle messages when context exceeds 30k tokens */
   get contextCompaction() {
-    return env("0SEC_FEATURE_CONTEXT_COMPACTION", true);
+    return env("ZERO_FEATURE_CONTEXT_COMPACTION", true);
   },
   /**
    * Re-send the opaque, model-bound Responses output item array on the next
    * turn. Default ON; set to 0 only for matched retained-reasoning A/B runs.
    */
   get retainedReasoning() {
-    return env("0SEC_FEATURE_RETAINED_REASONING", true);
+    return env("ZERO_FEATURE_RETAINED_REASONING", true);
   },
   /** Exploit script templates in shell prompt (blind SQLi, SSTI, auth chain) */
   get scriptTemplates() {
-    return env("0SEC_FEATURE_SCRIPT_TEMPLATES", true);
+    return env("ZERO_FEATURE_SCRIPT_TEMPLATES", true);
   },
   /** Dynamic vulnerability playbooks injected after recon phase */
   get dynamicPlaybooks() {
-    return env("0SEC_FEATURE_DYNAMIC_PLAYBOOKS", false);
+    return env("ZERO_FEATURE_DYNAMIC_PLAYBOOKS", false);
   },
   /** Just-in-time atomic DO/DON'T rules injected on a matching tool action */
   get ruleInjection() {
-    return env("0SEC_FEATURE_RULE_INJECTION", false);
+    return env("ZERO_FEATURE_RULE_INJECTION", false);
   },
   /** Agent writes plan/creds to disk, injected at reflection checkpoints */
   get externalMemory() {
-    return env("0SEC_FEATURE_EXTERNAL_MEMORY", false);
+    return env("ZERO_FEATURE_EXTERNAL_MEMORY", false);
   },
   /** Inject prior attempt findings when retrying (LLM-summarized progress handoff) */
   get progressHandoff() {
-    return env("0SEC_FEATURE_PROGRESS_HANDOFF", true);
+    return env("ZERO_FEATURE_PROGRESS_HANDOFF", true);
   },
   /** Allow the agent to search the web for CVE details, docs, and technique references */
   get webSearch() {
-    return env("0SEC_FEATURE_WEB_SEARCH", false);
+    return env("ZERO_FEATURE_WEB_SEARCH", false);
   },
   /** Interactive PTY sessions for exploits requiring interactivity (reverse shells, DB clients, SSH) */
   get ptySession() {
-    return env("0SEC_FEATURE_PTY_SESSION", false);
+    return env("ZERO_FEATURE_PTY_SESSION", false);
   },
   /**
    * Persistent, COMPUTE-ONLY Python REPL (`python_exec`, Phase-0). A framed
    * python3 kernel keeps state across calls for payload/parse/crypto/encode
    * work; networking is blocked at the socket source whenever an engagement is
-   * active. Default OFF — opt in via 0SEC_FEATURE_PYTHON_EXEC=1. Getter so
+   * active. Default OFF — opt in via ZERO_FEATURE_PYTHON_EXEC=1. Getter so
    * the CLI `--features` flag (set after this module is imported) is honored at
    * tool-dispatch time.
    */
   get pythonExec() {
-    return env("0SEC_FEATURE_PYTHON_EXEC", false);
+    return env("ZERO_FEATURE_PYTHON_EXEC", false);
   },
   /**
    * Expose the path-confined `analyze_binary` bridge to 0verse. Default OFF:
    * a model may request a long-running binary analysis only after an operator
-   * opts in with 0SEC_FEATURE_ZEROVERSE=1.
+   * opts in with ZERO_FEATURE_ZEROVERSE=1.
    */
   get zeroverse() {
-    return env("0SEC_FEATURE_ZEROVERSE", false);
+    return env("ZERO_FEATURE_ZEROVERSE", false);
   },
   /**
    * EGATS specialist routing (#557, HPTSA-inspired). When ON, an EGATS branch
@@ -146,22 +145,22 @@ var features = {
    * benchmark harness. Implemented as a getter so the CLI `--features` flag
    * (which sets the env var inside the command action, AFTER this module has
    * been imported) is honored at routing time. Enable via
-   * 0SEC_FEATURE_SPECIALIST_ROUTING=1.
+   * ZERO_FEATURE_SPECIALIST_ROUTING=1.
    */
   get specialistRouting() {
-    return env("0SEC_FEATURE_SPECIALIST_ROUTING", false);
+    return env("ZERO_FEATURE_SPECIALIST_ROUTING", false);
   },
   /** Self-consistency voting: run the structured verify pipeline N times and take the majority vote */
   get selfConsistencyVerify() {
-    return env("0SEC_FEATURE_CONSENSUS_VERIFY", false);
+    return env("ZERO_FEATURE_CONSENSUS_VERIFY", false);
   },
   /** Multi-modal agreement: cross-validate findings against foxguard (Rust pattern scanner) */
   get multiModalAgreement() {
-    return env("0SEC_FEATURE_MULTIMODAL", false);
+    return env("ZERO_FEATURE_MULTIMODAL", false);
   },
   /** Reachability gate: suppress findings whose sink is not reachable from an application entry point */
   get reachabilityGate() {
-    return env("0SEC_FEATURE_REACHABILITY_GATE", false);
+    return env("ZERO_FEATURE_REACHABILITY_GATE", false);
   },
   /**
    * Publishability / in-scope gate (issue #537 / #539). Decides
@@ -173,14 +172,14 @@ var features = {
    *
    * Default OFF: this gate can suppress reproducible findings, so it must be
    * explicitly opted into before any A/B claim. Disable/enable via
-   * 0SEC_FEATURE_PUBLISHABILITY_GATE.
+   * ZERO_FEATURE_PUBLISHABILITY_GATE.
    */
   get publishabilityGate() {
-    return env("0SEC_FEATURE_PUBLISHABILITY_GATE", false);
+    return env("ZERO_FEATURE_PUBLISHABILITY_GATE", false);
   },
   /** PoV gate: require a working, executable PoC per finding or downgrade to info */
   get povGate() {
-    return env("0SEC_FEATURE_POV_GATE", false);
+    return env("ZERO_FEATURE_POV_GATE", false);
   },
   /**
    * Intra-scan semantic dedupe post-pass (anchored incremental LLM
@@ -188,10 +187,10 @@ var features = {
    * Marks duplicates with a canonical mapping + cluster reason instead of
    * dropping them. Default OFF: it spends an LLM call per ≤50-finding batch
    * after the scan, so it must be explicitly opted into before any A/B
-   * claim. Toggle via 0SEC_FEATURE_SEMANTIC_DEDUPE.
+   * claim. Toggle via ZERO_FEATURE_SEMANTIC_DEDUPE.
    */
   get semanticDedupe() {
-    return env("0SEC_FEATURE_SEMANTIC_DEDUPE", false);
+    return env("ZERO_FEATURE_SEMANTIC_DEDUPE", false);
   },
   /**
    * Finding-specific remediation written by the model
@@ -207,33 +206,17 @@ var features = {
    * degrade cost, never correctness.
    */
   get llmRemediation() {
-    return env("0SEC_FEATURE_LLM_REMEDIATION", false);
-  },
-  /**
-   * Per-finding impact assessment (`assessImpact`, `triage/impact-assessment.ts`)
-   * written by the model: reachability tier, weaponizability, blast radius,
-   * business-impact tier. Default OFF — one extra LLM call per non-false-positive
-   * finding at report time.
-   *
-   * When on, the assessment feeds three things it is otherwise absent from:
-   * a real CVSS exploitability vector (AV/PR/UI from the reachability tier
-   * rather than the AV:N/severity-floor guess), the advisory's Impact +
-   * attack-prerequisites section, and the vendor-notification impact line. When
-   * off, all three fall back to today's category/severity heuristics — so this
-   * flag strictly adds fidelity, never changes the no-assessment output.
-   */
-  get impactAssessment() {
-    return env("0SEC_FEATURE_IMPACT_ASSESSMENT", false);
+    return env("ZERO_FEATURE_LLM_REMEDIATION", false);
   },
   /**
    * Incremental finding ranking post-pass (decimal-insertion between ranked
    * anchors, `triage/incremental-rank.ts`). Orders the report by comparative
    * promise (exploitability × impact × evidence strength). Default OFF: it
    * spends an LLM call per ≤50-finding batch; opt in before any A/B claim.
-   * Toggle via 0SEC_FEATURE_INCREMENTAL_RANK.
+   * Toggle via ZERO_FEATURE_INCREMENTAL_RANK.
    */
   get incrementalRank() {
-    return env("0SEC_FEATURE_INCREMENTAL_RANK", false);
+    return env("ZERO_FEATURE_INCREMENTAL_RANK", false);
   },
   /**
    * Static-finding PoC generation (#666 / EPIC #674 Part A). For findings that
@@ -247,10 +230,10 @@ var features = {
    *
    * Default OFF: it spends LLM + execution budget per static finding and must
    * be explicitly opted into before any A/B claim (A/B-able via the #656
-   * harness). Toggle via 0SEC_FEATURE_POC_GEN_STATIC.
+   * harness). Toggle via ZERO_FEATURE_POC_GEN_STATIC.
    */
   get pocGenStatic() {
-    return env("0SEC_FEATURE_POC_GEN_STATIC", false);
+    return env("ZERO_FEATURE_POC_GEN_STATIC", false);
   },
   /**
    * Inline validation / validate-on-save (#554). When ON, the native attack
@@ -268,10 +251,10 @@ var features = {
    * cost_per_flag claim. Implemented as a getter so the CLI `--features` flag
    * (which sets the env var inside the command action, AFTER this module is
    * imported) is honored at loop time. Enable via
-   * 0SEC_FEATURE_INLINE_VALIDATION=1.
+   * ZERO_FEATURE_INLINE_VALIDATION=1.
    */
   get inlineValidation() {
-    return env("0SEC_FEATURE_INLINE_VALIDATION", false);
+    return env("ZERO_FEATURE_INLINE_VALIDATION", false);
   },
   /**
    * WordPress plugin/theme fingerprinter + OSV CVE lookup.
@@ -286,7 +269,7 @@ var features = {
    * still honored at tool-dispatch time.
    */
   get wpFingerprint() {
-    return env("0SEC_FEATURE_WP_FINGERPRINT", true);
+    return env("ZERO_FEATURE_WP_FINGERPRINT", true);
   },
   /**
    * MongoDB ObjectID forge tool. Exposes the `mongo_objectid` tool to the
@@ -296,23 +279,23 @@ var features = {
    *
    * Default ON — this is a pure-computation utility with no network or
    * filesystem side effects, so there's no reason to gate it off. Disable
-   * via 0SEC_FEATURE_MONGO_OBJECTID_FORGE=0 or `--no-mongo-objectid-forge`
+   * via ZERO_FEATURE_MONGO_OBJECTID_FORGE=0 or `--no-mongo-objectid-forge`
    * for ablation. Implemented as a getter so the CLI `--features` flag
    * (which sets the env var inside the command action, AFTER this module
    * has been imported) is still honored at tool-dispatch time. Matches
    * the wpFingerprint pattern above. See packages/core/src/agent/objectid-forge.ts.
    */
   get mongoObjectIdForge() {
-    return env("0SEC_FEATURE_MONGO_OBJECTID_FORGE", true);
+    return env("ZERO_FEATURE_MONGO_OBJECTID_FORGE", true);
   },
   /**
-   * Live cloud-surface testing (0sec#925). Exposes `cloud_s3_probe` and
+   * Live cloud-surface testing (0#925). Exposes `cloud_s3_probe` and
    * `cloud_validate_credentials` to the attack agent so it can test S3 buckets
    * for public access + orphaned-bucket takeover and safely validate harvested
    * AWS credentials (read-only). All probes are anonymous or read/verify-only —
    * no writes, no data exfiltration beyond minimal proof.
    *
-   * Default OFF (opt-in via 0SEC_FEATURE_CLOUD_SURFACE=1). Probing a target
+   * Default OFF (opt-in via ZERO_FEATURE_CLOUD_SURFACE=1). Probing a target
    * org's bucket-name space or validating its harvested credentials is recon
    * AGAINST THAT ORG, so it is deny-by-default at two layers: this enablement
    * flag, AND an engagement-scope check in the tool handlers (a configured
@@ -323,19 +306,7 @@ var features = {
    * mongoObjectIdForge pattern above. See packages/core/src/agent/cloud-surface.ts.
    */
   get cloudSurface() {
-    return env("0SEC_FEATURE_CLOUD_SURFACE", false);
-  },
-  /**
-   * #978 (ADR-060) — agent fan-out. When ON, the agent gets the `start_scan`
-   * tool: it can dispatch CHILD scans (via the same POST /scans the UI uses)
-   * that run independently and report up the scan tree — the recursive
-   * sub-agent orchestration. Default OFF: fan-out multiplies scans/cost, so it
-   * stays opt-in even though the orchestrator enforces budget + a tree-level
-   * cap (max children/depth). Enable with 0SEC_FEATURE_AGENT_FANOUT=1.
-   * Getter so the CLI `--features` flag is honored at dispatch time.
-   */
-  get agentFanout() {
-    return env("0SEC_FEATURE_AGENT_FANOUT", false);
+    return env("ZERO_FEATURE_CLOUD_SURFACE", false);
   },
   // ── Phase-2 offensive-engine feature flags (dev-live-engine-recovery) ──
   // Each gates a tool that RUNS/BUILDS untrusted code or WEAPONIZES. They are
@@ -347,21 +318,21 @@ var features = {
   /**
    * `memsafety_fuzz` — clones/builds/fuzzes a source tree (sanitizer builds +
    * a fuzz harness), executing attacker-adjacent build scripts and native
-   * fuzz targets. Default OFF; opt in via 0SEC_FEATURE_MEMSAFETY=1. Building
+   * fuzz targets. Default OFF; opt in via ZERO_FEATURE_MEMSAFETY=1. Building
    * and running an untrusted tree is code execution, so it is deny-by-default
    * behind this flag AND an engagement scope.
    */
   get memsafetyFuzz() {
-    return env("0SEC_FEATURE_MEMSAFETY", false);
+    return env("ZERO_FEATURE_MEMSAFETY", false);
   },
   /**
    * `npm_dynamic_discovery` — installs and RUNS untrusted npm packages under
    * instrumentation to observe malicious install/runtime behaviour. Executing
    * arbitrary package code is the whole point, so it is deny-by-default behind
-   * this flag AND an engagement scope. Opt in via 0SEC_FEATURE_NPM_DISCOVERY=1.
+   * this flag AND an engagement scope. Opt in via ZERO_FEATURE_NPM_DISCOVERY=1.
    */
   get npmDynamicDiscovery() {
-    return env("0SEC_FEATURE_NPM_DISCOVERY", false);
+    return env("ZERO_FEATURE_NPM_DISCOVERY", false);
   },
   /**
    * `weaponize_kernel` — the kernel-exploit weaponization ladder. It only ever
@@ -369,19 +340,19 @@ var features = {
    * present. Highest-caution capability: deny-by-default behind this flag AND
    * an engagement scope AND a runtime artifact-presence check (the handler
    * refuses when the kernel-VM assets are absent). Opt in via
-   * 0SEC_FEATURE_KERNEL_WEAPONIZE=1.
+   * ZERO_FEATURE_KERNEL_WEAPONIZE=1.
    */
   get kernelWeaponize() {
-    return env("0SEC_FEATURE_KERNEL_WEAPONIZE", false);
+    return env("ZERO_FEATURE_KERNEL_WEAPONIZE", false);
   },
   /**
    * `cve_adapt` — adapts a public CVE PoC to the target and RUNS it to confirm
    * exploitability. Running an adapted exploit is code execution against the
    * target, so it is deny-by-default behind this flag AND an engagement scope.
-   * Opt in via 0SEC_FEATURE_CVE_ADAPT=1.
+   * Opt in via ZERO_FEATURE_CVE_ADAPT=1.
    */
   get cveAdapt() {
-    return env("0SEC_FEATURE_CVE_ADAPT", false);
+    return env("ZERO_FEATURE_CVE_ADAPT", false);
   },
   /**
    * Anti-honeypot flag-shape validator. When the agent calls the `done`
@@ -392,7 +363,7 @@ var features = {
    *
    * Default ON because legitimate flags pass the shape check trivially
    * and the false-positive rate on real flags should be near zero. Turn
-   * off via `0SEC_FEATURE_DECOY_DETECTION=0` or the CLI flag
+   * off via `ZERO_FEATURE_DECOY_DETECTION=0` or the CLI flag
    * `--no-decoy-detection` for ablation/testing.
    *
    * Implemented as a getter so the CLI flag (which flips the env var
@@ -402,7 +373,7 @@ var features = {
    * packages/core/src/agent/flag-validator.ts.
    */
   get decoyDetection() {
-    return env("0SEC_FEATURE_DECOY_DETECTION", true);
+    return env("ZERO_FEATURE_DECOY_DETECTION", true);
   },
   // ── Always-on triage filters (default ON, ablatable for A/B testing) ──
   /**
@@ -411,13 +382,13 @@ var features = {
    * sink names and rejects findings that look like "the function did its job".
    *
    * Default ON because that's the existing v0.6.0 behavior. Can be disabled
-   * via 0SEC_FEATURE_HOLDING_IT_WRONG=0 to test whether this filter is
+   * via ZERO_FEATURE_HOLDING_IT_WRONG=0 to test whether this filter is
    * suppressing real signal — the ceiling-analysis from 2026-04-06 identified
    * this as the strongest candidate for the unexplained XBOW finding-density
    * collapse from 14 → 4 between `features=none` and `features=all`.
    */
   get holdingItWrong() {
-    return env("0SEC_FEATURE_HOLDING_IT_WRONG", true);
+    return env("ZERO_FEATURE_HOLDING_IT_WRONG", true);
   },
   /**
    * `evidence_completeness <= 0.5` reject (`packages/core/src/agentic-scanner.ts:591`).
@@ -425,10 +396,10 @@ var features = {
    * gather enough cross-source evidence (request + response + analysis + ...).
    *
    * Default ON because that's the existing v0.6.0 behavior. Can be disabled
-   * via 0SEC_FEATURE_EVIDENCE_GATE=0 for ablation.
+   * via ZERO_FEATURE_EVIDENCE_GATE=0 for ablation.
    */
   get evidenceGate() {
-    return env("0SEC_FEATURE_EVIDENCE_GATE", true);
+    return env("ZERO_FEATURE_EVIDENCE_GATE", true);
   },
   /**
    * Learned per-finding triage router (`packages/core/src/triage/learned-router.ts`).
@@ -439,17 +410,17 @@ var features = {
    * the scan's slice type (xbow-wb, xbow-bb, npm).
    *
    * Default OFF until the router is validated via A/B testing on xbow-bench
-   * and npm-bench. See 0sec#113 for the design doc.
+   * and npm-bench. See 0#113 for the design doc.
    */
   get learnedRouter() {
-    return env("0SEC_FEATURE_LEARNED_ROUTER", false);
+    return env("ZERO_FEATURE_LEARNED_ROUTER", false);
   },
   /**
    * Dynamic per-finding triage routing (`packages/core/src/triage/router/`).
    * When enabled, every finding is sent through a `RouterModel` that
    * decides which subset of the 11 triage layers to invoke for that
    * specific finding. v0 ships an explicit-rule router encoded from the
-   * 0sec#72 per-profile ablation; a learned classifier replaces the
+   * 0#72 per-profile ablation; a learned classifier replaces the
    * rules in a follow-up PR without touching the dispatch site.
    *
    * Distinct from `learnedRouter` above: `learnedRouter` is the XGBoost
@@ -458,25 +429,11 @@ var features = {
    * the dispatch router gates which layers run AFTER the TP/FP score
    * model has spoken.
    *
-   * Default OFF — opt in via 0SEC_FEATURE_DYNAMIC_TRIAGE=1. See
-   * 0sec#113 for the design doc and 0sec#67 for the joint paper plan.
+   * Default OFF — opt in via ZERO_FEATURE_DYNAMIC_TRIAGE=1. See
+   * 0#113 for the design doc and 0#67 for the joint paper plan.
    */
   get dynamicTriageRouting() {
-    return env("0SEC_FEATURE_DYNAMIC_TRIAGE", false);
-  },
-  /**
-   * Opt-in cloud-sink webhook integration (`packages/core/src/cloud-sink.ts`).
-   * When enabled AND the user has set 0SEC_CLOUD_SINK + 0SEC_CLOUD_SCAN_ID,
-   * every finding and the final scan report are POSTed to the configured
-   * remote endpoint in real time.
-   *
-   * Default ON so the env-var trio is sufficient to enable streaming, but the
-   * flag exists so operators can force-disable the integration in environments
-   * where outbound HTTP from the scanner is not desired (e.g. air-gapped CI).
-   * Disable via 0SEC_FEATURE_CLOUD_SINK=0.
-   */
-  get cloudSink() {
-    return env("0SEC_FEATURE_CLOUD_SINK", true);
+    return env("ZERO_FEATURE_DYNAMIC_TRIAGE", false);
   },
   /**
    * Pre-recon CVE check (`packages/core/src/pre-recon-cve.ts`).
@@ -487,10 +444,10 @@ var features = {
    * where the agent has source access but no concrete leads.
    *
    * Default ON in white-box mode (no-op in black-box). Disable via
-   * 0SEC_FEATURE_PRE_RECON_CVE=0 for ablation.
+   * ZERO_FEATURE_PRE_RECON_CVE=0 for ablation.
    */
   get preReconCve() {
-    return env("0SEC_FEATURE_PRE_RECON_CVE", true);
+    return env("ZERO_FEATURE_PRE_RECON_CVE", true);
   },
   /**
    * Deterministic web-recon pre-pass (`packages/core/src/stages/web-recon-prepass.ts`).
@@ -501,25 +458,25 @@ var features = {
    * checks. It EMITS findings directly for what it can prove and injects a
    * "pursue these leads" block into the system prompt for what it can only hint.
    *
-   * Default ON (no-op in non-web modes). Gated behind 0SEC_FEATURE_WEB_RECON
+   * Default ON (no-op in non-web modes). Gated behind ZERO_FEATURE_WEB_RECON
    * so it can be disabled for ablation or offline runs. Implemented as a getter
    * so the CLI `--features` flag (which sets the env var inside the command
    * action, AFTER this module has been imported) is honored at stage time.
    */
   get webRecon() {
-    return env("0SEC_FEATURE_WEB_RECON", true);
+    return env("ZERO_FEATURE_WEB_RECON", true);
   },
   /**
    * Best-effort target-history preflight for source review. When a local repo
-   * path is known, 0sec infers repository/package/product hints, queries live
+   * path is known, 0 infers repository/package/product hints, queries live
    * prior-vulnerability intel, and injects a compact audit-graph summary into
    * the review prompt before the agent starts.
    *
    * Default ON for white-box/source-review modes. Disable via
-   * 0SEC_FEATURE_TARGET_HISTORY_PRESEED=0 for offline or ablation runs.
+   * ZERO_FEATURE_TARGET_HISTORY_PRESEED=0 for offline or ablation runs.
    */
   get targetHistoryPreseed() {
-    return env("0SEC_FEATURE_TARGET_HISTORY_PRESEED", true);
+    return env("ZERO_FEATURE_TARGET_HISTORY_PRESEED", true);
   },
   /**
    * Preserve credential / exploit-bearing messages verbatim during
@@ -534,15 +491,15 @@ var features = {
    * (a handful of extra messages preserved verbatim in the user
    * compaction-summary block) is small. BoxPwnr-inspired: see
    * `src/boxpwnr/solvers/single_loop_compactation.py` in 0ca/BoxPwnr,
-   * and 0sec#229 for the design discussion.
+   * and 0#229 for the design discussion.
    *
    * Implemented as a getter so the CLI `--features` flag — which sets
    * the env var inside the command action AFTER this module is imported
    * — is still honored at compaction time. Disable via
-   * 0SEC_FEATURE_PRESERVE_CRITICAL_MESSAGES=0 for ablation.
+   * ZERO_FEATURE_PRESERVE_CRITICAL_MESSAGES=0 for ablation.
    */
   get preserveCriticalMessages() {
-    return env("0SEC_FEATURE_PRESERVE_CRITICAL_MESSAGES", true);
+    return env("ZERO_FEATURE_PRESERVE_CRITICAL_MESSAGES", true);
   },
   /**
    * Two-stage budget-warning injection in the agent loop (#408).
@@ -558,14 +515,14 @@ var features = {
    * single short user-message injection at two specific turn boundaries,
    * and the win on long benchmarks (clean handoff instead of stray
    * exploration on the last turn) is well-documented in Strix's
-   * implementation. Disable via 0SEC_FEATURE_BUDGET_WARNINGS=0 for
+   * implementation. Disable via ZERO_FEATURE_BUDGET_WARNINGS=0 for
    * ablation. Implemented as a getter so the CLI `--features` flag —
    * which sets the env var inside the command action AFTER this module
    * is imported — is still honored at injection time (matches the
    * wpFingerprint / preserveCriticalMessages pattern).
    */
   get budgetWarnings() {
-    return env("0SEC_FEATURE_BUDGET_WARNINGS", true);
+    return env("ZERO_FEATURE_BUDGET_WARNINGS", true);
   },
   /**
    * Per-file orchestration for the research and audit stages (#285).
@@ -579,14 +536,14 @@ var features = {
    * Trade-off: total token spend grows roughly N × per-file budget instead
    * of capped at a single session's budget. For a 50-file package, that
    * could be a 5-10× cost increase on research. Disable via
-   * `0SEC_FEATURE_PER_ITEM_ORCHESTRATION=0` to revert to the shared-session
+   * `ZERO_FEATURE_PER_ITEM_ORCHESTRATION=0` to revert to the shared-session
    * behavior — useful for cost-bounded benchmarks.
    *
    * Implemented as a getter so the env var is honored at orchestration time
    * (matches the wpFingerprint / mongoObjectIdForge pattern).
    */
   get perItemOrchestration() {
-    return env("0SEC_FEATURE_PER_ITEM_ORCHESTRATION", true);
+    return env("ZERO_FEATURE_PER_ITEM_ORCHESTRATION", true);
   },
   /**
    * JIT skill loading (`packages/core/src/agent/skills/`).
@@ -595,20 +552,21 @@ var features = {
    * them into working context mid-scan. Skills replace the monolithic
    * playbook injection with targeted, on-demand knowledge (#410, #457).
    *
-   * Default OFF until the skill registry is validated via A/B testing.
+   * Default OFF unless an assigned cloud methodology manifest opts this run in.
+   * An explicit feature flag still takes precedence over that default.
    * Implemented as a getter so the CLI `--features` flag — which sets
    * the env var inside the command action, AFTER this module has been
    * imported — is still honored at tool-dispatch time.
    */
   get jitSkills() {
-    return env("0SEC_FEATURE_JIT_SKILLS", false);
+    return env("ZERO_FEATURE_JIT_SKILLS", Boolean(process.env["ZERO_AUDIT_SKILLS_MANIFEST"]?.trim()));
   },
   /**
    * Execution-journal shadow mode (#494, first additive slice).
    *
    * When ON, the live agent loop ALSO writes append-only journal entries
    * (`tool_call`, `tool_result`, `finding`, `done`) to
-   * `~/.0sec/runs/<scanId>/journal.jsonl` as it runs — a durable,
+   * `~/.0/runs/<scanId>/journal.jsonl` as it runs — a durable,
    * replayable trace alongside the existing in-memory conversation window.
    * This is strictly additive: the loop continues to drive off its own
    * conversation state, the journal is write-only here, and a failed
@@ -621,11 +579,11 @@ var features = {
    * moat-ablation harness before any A/B claim. Implemented as a getter so
    * the CLI `--features` flag (which sets the env var inside the command
    * action, AFTER this module has been imported) is honored at loop time.
-   * Enable via 0SEC_FEATURE_EXECUTION_JOURNAL=1 or `--features
+   * Enable via ZERO_FEATURE_EXECUTION_JOURNAL=1 or `--features
    * execution-journal`.
    */
   get executionJournal() {
-    return env("0SEC_FEATURE_EXECUTION_JOURNAL", false);
+    return env("ZERO_FEATURE_EXECUTION_JOURNAL", false);
   },
   /**
    * Execution-journal context routing (#494, slice 2).
@@ -641,7 +599,7 @@ var features = {
    * Independent of `executionJournal` (the shadow-WRITE flag) on purpose so
    * the moat-ablation harness can toggle write and route separately for a
    * clean A/B. Rehydrate is a READER, though, so it only does anything when
-   * a journal was written for the run — it reads `~/.0sec/runs/<scanId>/
+   * a journal was written for the run — it reads `~/.0/runs/<scanId>/
    * journal.jsonl` regardless of how it got there (shadow mode this slice,
    * or specialists in a later slice). When the journal is missing, empty, or
    * corrupt the loop falls back to the existing DB-blob / fresh-prompt
@@ -654,11 +612,11 @@ var features = {
    * must be explicitly opted into before any A/B claim. Implemented as a
    * getter so the CLI `--features` flag (which sets the env var inside the
    * command action, AFTER this module has been imported) is honored at loop
-   * time. Enable via 0SEC_FEATURE_JOURNAL_REHYDRATE=1 or `--features
+   * time. Enable via ZERO_FEATURE_JOURNAL_REHYDRATE=1 or `--features
    * journal-rehydrate`.
    */
   get journalRehydrate() {
-    return env("0SEC_FEATURE_JOURNAL_REHYDRATE", false);
+    return env("ZERO_FEATURE_JOURNAL_REHYDRATE", false);
   },
   /**
    * Loot / foothold ledger for opportunistic exploit chaining (#567).
@@ -678,14 +636,14 @@ var features = {
    * tool), matches the `preserveCriticalMessages` rationale — recovering a
    * credential in turn 12 that's needed in turn 38 is a large win on long-tail
    * challenges — and the cost (a short, size-capped block per turn) is small.
-   * Disable via 0SEC_FEATURE_LOOT_LEDGER=0 or `--no-loot-ledger` for
+   * Disable via ZERO_FEATURE_LOOT_LEDGER=0 or `--no-loot-ledger` for
    * ablation. Implemented as a getter so the CLI `--features` flag (which sets
    * the env var inside the command action, AFTER this module has been
    * imported) is honored at tool-dispatch / injection time — matches the
    * wpFingerprint / preserveCriticalMessages pattern.
    */
   get lootLedger() {
-    return env("0SEC_FEATURE_LOOT_LEDGER", true);
+    return env("ZERO_FEATURE_LOOT_LEDGER", true);
   },
   /**
    * Typed TODO / plan ledger (`packages/core/src/agent/task-ledger.ts`).
@@ -704,13 +662,13 @@ var features = {
    * failure mode of an unused tool is a few hundred wasted schema tokens
    * rather than wrong behavior. Note for whoever publishes benchmark numbers
    * next: this DOES change the default tool list, so re-baseline before
-   * quoting a figure across this change. Disable via 0SEC_FEATURE_AGENT_PLAN=0
+   * quoting a figure across this change. Disable via ZERO_FEATURE_AGENT_PLAN=0
    * or `--features no-agent-plan` for ablation. Getter so the CLI `--features`
    * flag (which sets the env var AFTER this module is imported) is honored at
    * tool-dispatch time.
    */
   get agentPlan() {
-    return env("0SEC_FEATURE_AGENT_PLAN", false);
+    return env("ZERO_FEATURE_AGENT_PLAN", false);
   },
   /**
    * Task-drift detection (`packages/core/src/agent/drift.ts`).
@@ -733,10 +691,10 @@ var features = {
    * to a newly-discovered lead is lexically indistinguishable from a derail.
    * Repo convention is explicit that behavior-steering features stay opt-in
    * until A/B'd, and this is squarely one. Enable via
-   * 0SEC_FEATURE_DRIFT_DETECTION=1 or `--features drift-detection`.
+   * ZERO_FEATURE_DRIFT_DETECTION=1 or `--features drift-detection`.
    */
   get driftDetection() {
-    return env("0SEC_FEATURE_DRIFT_DETECTION", false);
+    return env("ZERO_FEATURE_DRIFT_DETECTION", false);
   },
   /**
    * OAST out-of-band interaction collaborator + oracle (#659).
@@ -748,12 +706,12 @@ var features = {
    * evidence and feeds the loot ledger.
    *
    * Default OFF — the tools are inert without a deployed collaborator. Enable
-   * with 0SEC_FEATURE_OAST=1 AND point 0SEC_OAST_URL at the self-hosted
+   * with ZERO_FEATURE_OAST=1 AND point ZERO_OAST_URL at the self-hosted
    * collaborator server (see packages/core/src/oast/server.ts). Getter (not a
    * const) so the CLI `--features` flag is honored at tool-dispatch time.
    */
   get oastCollaborator() {
-    return env("0SEC_FEATURE_OAST", false);
+    return env("ZERO_FEATURE_OAST", false);
   },
   /**
    * Anthropic prompt caching (`cache_control: {type: "ephemeral"}`) over the
@@ -776,17 +734,17 @@ var features = {
    * never see an Anthropic-shaped field regardless of this flag (see
    * `providerSupportsPromptCache`).
    *
-   * Disable via 0SEC_FEATURE_PROMPT_CACHE=0 — worth doing only to isolate a
+   * Disable via ZERO_FEATURE_PROMPT_CACHE=0 — worth doing only to isolate a
    * suspected provider-side caching bug, or to measure the uncached baseline.
    * Implemented as a getter so a late env mutation (CLI `--features`, which
    * runs after this module is imported) is honoured at request-build time.
    */
   get promptCache() {
-    return env("0SEC_FEATURE_PROMPT_CACHE", true);
+    return env("ZERO_FEATURE_PROMPT_CACHE", true);
   }
 };
 function presetRaisesDefault(key) {
-  const raw = process.env["0SEC_FEATURE_PRESET"];
+  const raw = process.env["ZERO_FEATURE_PRESET"];
   if (!raw)
     return false;
   const preset = resolveFeaturePreset(raw);
@@ -884,7 +842,7 @@ function sanitizeFields(input) {
 var LEVEL_RANK = { info: 10, warn: 20, error: 30 };
 var OFF_RANK = Number.POSITIVE_INFINITY;
 function minimumRank() {
-  const raw = process.env["0SEC_DIAG_LEVEL"];
+  const raw = process.env["ZERO_DIAG_LEVEL"];
   if (!raw)
     return LEVEL_RANK.info;
   switch (raw.trim().toLowerCase()) {
@@ -904,9 +862,9 @@ function minimumRank() {
 function formatDiagnosticLine(event) {
   const keys = Object.keys(event.fields);
   if (keys.length === 0)
-    return `[0sec] ${event.message}`;
+    return `[0] ${event.message}`;
   const rendered = keys.map((k) => `${k}=${event.fields[k]}`).join(" ");
-  return `[0sec] ${event.message} (${rendered})`;
+  return `[0] ${event.message} (${rendered})`;
 }
 var stderrDiagnosticSink = {
   emit(event) {
@@ -1065,295 +1023,11 @@ function _resetDiagnosticsForTests() {
   diag.resetForTests();
 }
 
-// packages/core/dist/cloud/credentials.js
-import { readFileSync, statSync } from "node:fs";
-import { join } from "node:path";
-var DEFAULT_CLOUD_HOST = "https://cloud.0.security";
-var CloudAuthMissingError = class extends Error {
-  constructor(message) {
-    super(message);
-    this.name = "CloudAuthMissingError";
-  }
-};
-var CloudAuthError = class extends Error {
-  status;
-  constructor(message, status) {
-    super(message);
-    this.status = status;
-    this.name = "CloudAuthError";
-  }
-};
-function loadCloudCredentials(opts = {}) {
-  const env2 = opts.env ?? process.env;
-  const warn = opts.warn ?? ((m) => process.stderr.write(`${m}
-`));
-  const envTok = env2["0SEC_CLOUD_TOKEN"]?.trim();
-  if (envTok) {
-    const envHost = normaliseHost(env2["0SEC_CLOUD_HOST"]?.trim() ?? DEFAULT_CLOUD_HOST);
-    return { host: envHost, token: envTok, source: "env" };
-  }
-  const path = join(homeStateDir(opts.homeDir), "cloud.env");
-  let raw;
-  try {
-    raw = readFileSync(path, "utf-8");
-  } catch (err) {
-    const code = err.code;
-    if (code === "ENOENT") {
-      throw new CloudAuthMissingError(`0sec-cloud credentials not found. Run \`0sec auth login\` or set 0SEC_CLOUD_TOKEN in env, or create ${path} (chmod 600) with 0SEC_CLOUD_TOKEN=\u2026 (optionally 0SEC_CLOUD_HOST=\u2026).`);
-    }
-    throw err;
-  }
-  try {
-    const st = statSync(path);
-    const mode = st.mode & 511;
-    if (mode !== 384) {
-      warn(`[0sec cloud] WARNING: ${path} mode is ${mode.toString(8).padStart(3, "0")} (expected 600). Run: chmod 600 ${path}`);
-    }
-  } catch {
-  }
-  const parsed = parseEnvFile(raw);
-  const fileTok = parsed["0SEC_CLOUD_TOKEN"]?.trim();
-  if (!fileTok) {
-    throw new CloudAuthMissingError(`0sec-cloud credentials in ${path} are incomplete: 0SEC_CLOUD_TOKEN is required.`);
-  }
-  const fileHost = normaliseHost(parsed["0SEC_CLOUD_HOST"]?.trim() ?? DEFAULT_CLOUD_HOST);
-  return { host: fileHost, token: fileTok, source: "file" };
-}
-function normaliseHost(host) {
-  let h = host;
-  if (!/^https?:\/\//.test(h)) {
-    throw new CloudAuthMissingError(`0SEC_CLOUD_HOST must be an http(s) URL (got ${JSON.stringify(host)}).`);
-  }
-  while (h.endsWith("/"))
-    h = h.slice(0, -1);
-  return h;
-}
-function parseEnvFile(raw) {
-  const out = {};
-  const lines = raw.split(/\r?\n/);
-  for (let i = 0; i < lines.length; i++) {
-    const line = lines[i];
-    const trimmed = line.trim();
-    if (trimmed.length === 0)
-      continue;
-    if (trimmed.startsWith("#"))
-      continue;
-    const eq = trimmed.indexOf("=");
-    if (eq <= 0) {
-      throw new CloudAuthMissingError(`Malformed cloud.env at line ${i + 1}: expected KEY=VALUE, got ${JSON.stringify(line)}`);
-    }
-    const key = trimmed.slice(0, eq).trim();
-    const value = trimmed.slice(eq + 1).trim();
-    if (!/^[A-Z0-9_]+$/.test(key)) {
-      throw new CloudAuthMissingError(`Malformed cloud.env at line ${i + 1}: invalid key ${JSON.stringify(key)}`);
-    }
-    out[key] = value;
-  }
-  return out;
-}
-
-// packages/core/dist/cloud/client.js
-var CloudError = class extends Error {
-  status;
-  path;
-  code;
-  constructor(message, status, path, code) {
-    super(message);
-    this.status = status;
-    this.path = path;
-    this.code = code;
-    this.name = "CloudError";
-  }
-};
-var CloudUnauthorizedError = class extends CloudError {
-  constructor(path) {
-    super(`0sec-cloud auth rejected (HTTP 401) on ${path}. Run \`0sec auth login\` to refresh.`, 401, path);
-    this.name = "CloudUnauthorizedError";
-  }
-};
-var CloudForbiddenError = class extends CloudError {
-  constructor(path) {
-    super(`0sec-cloud forbidden (HTTP 403) on ${path}. Token lacks scope for this resource.`, 403, path);
-    this.name = "CloudForbiddenError";
-  }
-};
-var CloudNetworkError = class extends CloudError {
-  constructor(message, path) {
-    super(`0sec-cloud network error on ${path}: ${message}`, void 0, path);
-    this.name = "CloudNetworkError";
-  }
-};
-function healthPath(host) {
-  try {
-    const hostname = new URL(host).hostname.toLowerCase();
-    if (hostname === "cloud.0sec.ai" || hostname === "cloud.0.security") {
-      return "/api/health";
-    }
-  } catch {
-  }
-  return "/health";
-}
-var CloudClient = class {
-  host;
-  token;
-  fetchImpl;
-  constructor(opts) {
-    this.host = opts.host;
-    this.token = opts.token;
-    this.fetchImpl = opts.fetchImpl ?? fetch;
-  }
-  /**
-   * Verify cloud reachability through its health route. The hosted dashboard
-   * uses `/api/health`; a self-hosted receiver uses `/health`.
-   */
-  async pingHealth() {
-    return this.getJson(healthPath(this.host));
-  }
-  /**
-   * Fetch the hosted inference model catalog — available models, pricing,
-   * wire API protocol, and context limits. Used at runtime for model
-   * selection and by the hosted provider to determine per-model capabilities.
-   * Returns the raw list response; the caller caches/filters as needed.
-   */
-  async getInferenceModels() {
-    return this.getJson("/api/inference/v1/models");
-  }
-  /**
-   * Fetch the organization's hosted inference credit availability. The service
-   * calculates the percentage from Autumn's current pool; holds reduce availability.
-   * Older gateways without percentage metadata remain explicitly unavailable.
-   */
-  async getInferenceAccount() {
-    const account = await this.getJson("/api/inference/account");
-    const credits = account.credits;
-    if (!credits || typeof credits !== "object" || typeof credits.featureId !== "string" || !credits.featureId.trim() || typeof credits.remaining !== "number" || !Number.isFinite(credits.remaining) || credits.remaining < 0 || credits.granted !== null && (typeof credits.granted !== "number" || !Number.isFinite(credits.granted) || credits.granted < 0)) {
-      return { ...account, credits: null };
-    }
-    const remainingPercent = typeof credits.remainingPercent === "number" && Number.isFinite(credits.remainingPercent) && credits.remainingPercent >= 0 && credits.remainingPercent <= 100 && credits.granted !== null && credits.granted > 0 && credits.remaining <= credits.granted ? credits.remainingPercent : null;
-    const nextResetAt = typeof credits.nextResetAt === "number" && Number.isSafeInteger(credits.nextResetAt) && credits.nextResetAt > 0 && credits.nextResetAt <= 864e13 ? credits.nextResetAt : null;
-    return { ...account, credits: {
-      featureId: credits.featureId,
-      granted: credits.granted,
-      remaining: credits.remaining,
-      remainingPercent,
-      nextResetAt
-    } };
-  }
-  /**
-   * Fetch request-level usage metadata for the operator's hosted
-   * inference sessions. Returns lightweight metadata records (model,
-   * tokens, provider, timestamp) — no prompt/response payload.
-   */
-  async getInferenceUsage() {
-    return this.getJson("/api/inference/usage");
-  }
-  /**
-   * Generic JSON POST helper with the same error mapping as getJson.
-   * Used by `0sec connect` to enqueue scans and schedules.
-   */
-  async postJson(path, body) {
-    const url = `${this.host}${path}`;
-    let res;
-    try {
-      res = await this.fetchImpl(url, {
-        method: "POST",
-        headers: { ...this.headers(), "Content-Type": "application/json" },
-        body: JSON.stringify(body)
-      });
-    } catch (err) {
-      const msg = err instanceof Error ? err.message : String(err);
-      throw new CloudNetworkError(this.scrub(msg), path);
-    }
-    if (!res.ok) {
-      let code;
-      try {
-        const parsed = await res.json();
-        const raw = typeof parsed?.error === "object" ? parsed.error?.code : void 0;
-        if (typeof raw === "string" && raw.length > 0)
-          code = raw;
-      } catch {
-      }
-      this.throwForStatus(res.status, path, code);
-    }
-    return await res.json();
-  }
-  /**
-   * Generic JSON GET helper. Public so future modules (scans, findings)
-   * can reuse the same error mapping without duplicating it. Not exported
-   * past the package boundary — see ./index.ts.
-   */
-  async getJson(path) {
-    const url = `${this.host}${path}`;
-    let res;
-    try {
-      res = await this.fetchImpl(url, {
-        method: "GET",
-        headers: this.headers()
-      });
-    } catch (err) {
-      const msg = err instanceof Error ? err.message : String(err);
-      throw new CloudNetworkError(this.scrub(msg), path);
-    }
-    if (!res.ok) {
-      let code;
-      try {
-        const body = await res.json();
-        const raw = body?.error?.code;
-        if (typeof raw === "string" && raw.length > 0)
-          code = raw;
-      } catch {
-      }
-      this.throwForStatus(res.status, path, code);
-    }
-    return await res.json();
-  }
-  /** Throw a typed error for a non-2xx status, carrying the gateway's body `code`. */
-  throwForStatus(status, path, code) {
-    if (status === 401)
-      throw new CloudUnauthorizedError(path);
-    if (status === 403)
-      throw new CloudForbiddenError(path);
-    throw new CloudError(`0sec-cloud request failed (HTTP ${status}${code ? ` ${code}` : ""}) on ${path}.`, status, path, code);
-  }
-  /**
-   * Throw a typed error for non-2xx responses. Public so direct callers
-   * (e.g. an integration test driving raw fetch) can reuse the mapping.
-   */
-  assertOk(res, path) {
-    if (res.ok)
-      return;
-    if (res.status === 401)
-      throw new CloudUnauthorizedError(path);
-    if (res.status === 403)
-      throw new CloudForbiddenError(path);
-    throw new CloudError(`0sec-cloud request failed (HTTP ${res.status}) on ${path}.`, res.status, path);
-  }
-  // ── internals ──
-  headers() {
-    return {
-      Authorization: `Bearer ${this.token}`,
-      Accept: "application/json",
-      "User-Agent": `0sec-cli/${VERSION}`
-    };
-  }
-  /**
-   * Strip anything that looks like our own token from a string. The
-   * cloud token may be interpolated into a TLS-layer error message in
-   * exotic failure modes — we redact it to keep the no-leak invariant
-   * local to this module.
-   */
-  scrub(s) {
-    if (!this.token)
-      return s;
-    return s.split(this.token).join("[REDACTED]");
-  }
-};
-
 // packages/core/dist/runtime/llm-api.js
-import { randomUUID } from "node:crypto";
-import { appendFileSync, existsSync, readFileSync as readFileSync2, renameSync, writeFileSync } from "node:fs";
+import { createHash, randomUUID } from "node:crypto";
+import { appendFileSync, existsSync, readFileSync, renameSync, writeFileSync } from "node:fs";
 import { homedir } from "node:os";
-import { join as join2 } from "node:path";
+import { join } from "node:path";
 
 // packages/core/dist/runtime/prompt-cache.js
 var MAX_CACHE_BREAKPOINTS = 4;
@@ -1369,7 +1043,7 @@ function providerSupportsPromptCache(provider) {
   return readExtraCacheProviders().has(provider);
 }
 function readExtraCacheProviders() {
-  const raw = process.env["0SEC_PROMPT_CACHE_EXTRA_PROVIDERS"];
+  const raw = process.env["ZERO_PROMPT_CACHE_EXTRA_PROVIDERS"];
   if (!raw)
     return /* @__PURE__ */ new Set();
   return new Set(raw.split(",").map((entry) => entry.trim().toLowerCase()).filter((entry) => entry.length > 0));
@@ -1438,7 +1112,7 @@ function isWireBlockArray(blocks) {
 }
 var azureRegionCache = /* @__PURE__ */ new Map();
 async function probeAzureRegion(baseUrl, apiKey, fetchImpl = fetch) {
-  const override = process.env["0SEC_REGION_OVERRIDE"];
+  const override = process.env["ZERO_REGION_OVERRIDE"];
   if (override && override.trim().length > 0) {
     return override.trim();
   }
@@ -1500,7 +1174,7 @@ function prettyRegion(code) {
   };
   return map[code.toLowerCase()] ?? code;
 }
-var PROVIDER_BANNER_KEY = /* @__PURE__ */ Symbol.for("0sec.core.loggedProviderStartup");
+var PROVIDER_BANNER_KEY = /* @__PURE__ */ Symbol.for("0.core.loggedProviderStartup");
 var loggedProviderStartup = (() => {
   const g = globalThis;
   if (!g[PROVIDER_BANNER_KEY])
@@ -1508,7 +1182,7 @@ var loggedProviderStartup = (() => {
   return g[PROVIDER_BANNER_KEY];
 })();
 function appendNativeTrace(record) {
-  const file = process.env["0SEC_TRACE_NATIVE_RESPONSES"];
+  const file = process.env["ZERO_TRACE_NATIVE_RESPONSES"];
   if (!file)
     return;
   try {
@@ -1518,7 +1192,7 @@ function appendNativeTrace(record) {
   }
 }
 function shouldLogProviderStartup() {
-  return process.env["0SEC_SUPPRESS_PROVIDER_STARTUP_LOG"] !== "1";
+  return process.env["ZERO_SUPPRESS_PROVIDER_STARTUP_LOG"] !== "1";
 }
 function isRetryableHttpStatus(status) {
   return status === 429 || status === 500 || status === 502 || status === 503 || status === 504;
@@ -1528,7 +1202,7 @@ var TRANSIENT_STREAM_ERROR_PATTERNS = [
   "response stream failed"
 ];
 function llmStreamMaxAttempts() {
-  const raw = process.env["0SEC_LLM_STREAM_MAX_ATTEMPTS"];
+  const raw = process.env["ZERO_LLM_STREAM_MAX_ATTEMPTS"];
   if (raw == null || raw.trim() === "")
     return 3;
   const n = Number.parseInt(raw, 10);
@@ -1575,28 +1249,28 @@ function isRetryableTransportCode(code) {
   ].includes(code);
 }
 function llmMaxRetries() {
-  const raw = process.env["0SEC_LLM_MAX_RETRIES"];
+  const raw = process.env["ZERO_LLM_MAX_RETRIES"];
   if (raw == null || raw.trim() === "")
     return 6;
   const n = Number.parseInt(raw, 10);
   return Number.isFinite(n) && n >= 0 ? n : 6;
 }
 function llmMaxRetryWaitMs() {
-  const raw = process.env["0SEC_LLM_MAX_RETRY_WAIT_MS"];
+  const raw = process.env["ZERO_LLM_MAX_RETRY_WAIT_MS"];
   if (raw == null || raw.trim() === "")
     return 6e4;
   const n = Number.parseInt(raw, 10);
   return Number.isFinite(n) && n > 0 ? n : 6e4;
 }
 function llm429MaxRetries() {
-  const raw = process.env["0SEC_LLM_429_MAX_RETRIES"] ?? process.env["0SEC_LLM_MAX_RETRIES"];
+  const raw = process.env["ZERO_LLM_429_MAX_RETRIES"] ?? process.env["ZERO_LLM_MAX_RETRIES"];
   if (raw == null || raw.trim() === "")
     return 12;
   const n = Number.parseInt(raw, 10);
   return Number.isFinite(n) && n >= 0 ? n : 12;
 }
 function llm429MaxRetryWaitMs() {
-  const raw = process.env["0SEC_LLM_429_MAX_RETRY_WAIT_MS"] ?? process.env["0SEC_LLM_MAX_RETRY_WAIT_MS"];
+  const raw = process.env["ZERO_LLM_429_MAX_RETRY_WAIT_MS"] ?? process.env["ZERO_LLM_MAX_RETRY_WAIT_MS"];
   if (raw == null || raw.trim() === "")
     return 3e5;
   const n = Number.parseInt(raw, 10);
@@ -1707,14 +1381,14 @@ function parseUsageLimitReached(body) {
   return details;
 }
 function llmStreamIdleTimeoutMs() {
-  const raw = process.env["0SEC_LLM_STREAM_IDLE_TIMEOUT_MS"];
+  const raw = process.env["ZERO_LLM_STREAM_IDLE_TIMEOUT_MS"];
   if (raw == null || raw.trim() === "")
     return 12e4;
   const n = Number.parseInt(raw, 10);
   return Number.isFinite(n) && n > 0 ? n : 12e4;
 }
 function llmStreamEventIdleTimeoutMs() {
-  const raw = process.env["0SEC_LLM_STREAM_EVENT_IDLE_TIMEOUT_MS"];
+  const raw = process.env["ZERO_LLM_STREAM_EVENT_IDLE_TIMEOUT_MS"];
   if (raw == null || raw.trim() === "")
     return 24e4;
   const n = Number.parseInt(raw, 10);
@@ -1864,7 +1538,7 @@ var AZURE_FOUNDRY_DEPLOYMENT_IDS = {
   "gpt-5.6-terra": true
 };
 function parseLlmFallbackChain(env2 = process.env) {
-  const raw = env2["0SEC_LLM_FALLBACK"];
+  const raw = env2["ZERO_LLM_FALLBACK"];
   if (!raw || raw.trim().length === 0)
     return [];
   const entries = [];
@@ -1881,8 +1555,7 @@ function parseLlmFallbackChain(env2 = process.env) {
     xai: true,
     opencode: true,
     copilot: true,
-    google: true,
-    hosted: true
+    google: true
   };
   for (const part of raw.split(",")) {
     const trimmed = part.trim();
@@ -1890,17 +1563,17 @@ function parseLlmFallbackChain(env2 = process.env) {
       continue;
     const colonIdx = trimmed.indexOf(":");
     if (colonIdx < 1 || colonIdx === trimmed.length - 1) {
-      diag.warn("fallback_chain_malformed_entry", `0SEC_LLM_FALLBACK: malformed entry "${trimmed}" (expected provider:model)`, { entry: trimmed, expected: "provider:model" });
+      diag.warn("fallback_chain_malformed_entry", `ZERO_LLM_FALLBACK: malformed entry "${trimmed}" (expected provider:model)`, { entry: trimmed, expected: "provider:model" });
       continue;
     }
     const provider = trimmed.slice(0, colonIdx);
     const model = trimmed.slice(colonIdx + 1).trim();
     if (!VALID_PROVIDERS[provider]) {
-      diag.warn("fallback_chain_unknown_provider", `0SEC_LLM_FALLBACK: unknown provider "${provider}" in "${trimmed}"`, { entry: trimmed, provider });
+      diag.warn("fallback_chain_unknown_provider", `ZERO_LLM_FALLBACK: unknown provider "${provider}" in "${trimmed}"`, { entry: trimmed, provider });
       continue;
     }
     if (!model) {
-      diag.warn("fallback_chain_empty_model", `0SEC_LLM_FALLBACK: empty model in "${trimmed}"`, { entry: trimmed, provider });
+      diag.warn("fallback_chain_empty_model", `ZERO_LLM_FALLBACK: empty model in "${trimmed}"`, { entry: trimmed, provider });
       continue;
     }
     entries.push({ provider, model });
@@ -1943,7 +1616,7 @@ function resolveFailoverProvider(provider, model, env2 = process.env, apiKey) {
       return { apiKey: key, baseUrl: env2.ANTHROPIC_BASE_URL ?? "https://api.anthropic.com", wireApi: "chat_completions" };
     }
     case "chatgpt-codex": {
-      if (!env2["0SEC_CHATGPT_ACCESS_TOKEN"] && !env2["0SEC_CHATGPT_OAUTH_REFRESH_TOKEN"] && !readChatGptCodexAuthFile(env2))
+      if (!env2["ZERO_CHATGPT_ACCESS_TOKEN"] && !env2["ZERO_CHATGPT_OAUTH_REFRESH_TOKEN"] && !readChatGptCodexAuthFile(env2))
         return void 0;
       return { apiKey: "", baseUrl: CODEX_API_ENDPOINT, wireApi: "responses" };
     }
@@ -1978,39 +1651,21 @@ function resolveFailoverProvider(provider, model, env2 = process.env, apiKey) {
       return { apiKey: key, baseUrl: env2.OPENCODE_BASE_URL ?? OPENCODE_DEFAULT_BASE_URL, wireApi: opencodeWireApiForModel(model) };
     }
     case "copilot": {
-      const key = apiKey ?? env2["0SEC_COPILOT_GITHUB_TOKEN"];
+      const key = apiKey ?? env2["ZERO_COPILOT_GITHUB_TOKEN"];
       if (!key)
         return void 0;
       return { apiKey: key, baseUrl: env2.COPILOT_BASE_URL ?? COPILOT_API_BASE, wireApi: "chat_completions" };
     }
     case "google": {
-      if (!env2["0SEC_GEMINI_ACCESS_TOKEN"] && !env2["0SEC_GEMINI_OAUTH_REFRESH_TOKEN"])
+      if (!env2["ZERO_GEMINI_ACCESS_TOKEN"] && !env2["ZERO_GEMINI_OAUTH_REFRESH_TOKEN"])
         return void 0;
       return { apiKey: "", baseUrl: CODE_ASSIST_ENDPOINT, wireApi: "google_generate_content" };
     }
-    case "hosted": {
-      try {
-        const creds = loadCloudCredentials({
-          env: env2,
-          warn: () => {
-          }
-        });
-        return {
-          apiKey: creds.token,
-          baseUrl: `${creds.host}/api/inference/v1`,
-          wireApi: "chat_completions"
-        };
-      } catch (err) {
-        if (err instanceof CloudAuthMissingError)
-          return void 0;
-        throw err;
-      }
-    }
   }
 }
 var fallbackChainCache;
 function getFallbackChain(env2) {
-  const raw = env2["0SEC_LLM_FALLBACK"];
+  const raw = env2["ZERO_LLM_FALLBACK"];
   if (!fallbackChainCache || fallbackChainCache.raw !== raw) {
     fallbackChainCache = { raw, entries: parseLlmFallbackChain(env2) };
   }
@@ -2020,7 +1675,7 @@ var ZAI_DEFAULT_BASE_URL = "https://api.z.ai/api/anthropic";
 var ZAI_DEFAULT_MODEL = "glm-5.3";
 var ZAI_DEFAULT_THINKING_BUDGET = 2048;
 function zaiThinkingBudget() {
-  const raw = process.env["0SEC_ZAI_THINKING_BUDGET"];
+  const raw = process.env["ZERO_ZAI_THINKING_BUDGET"];
   if (raw == null || raw.trim().length === 0)
     return ZAI_DEFAULT_THINKING_BUDGET;
   const n = Number.parseInt(raw, 10);
@@ -2031,22 +1686,28 @@ var KIMI_DEFAULT_MODEL = "k3";
 var QWEN_DEFAULT_BASE_URL = "https://token-plan.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1";
 var QWEN_DEFAULT_MODEL = "qwen3.8-max";
 var CODEX_API_ENDPOINT = "https://chatgpt.com/backend-api/codex/responses";
+var CODEX_CLIENT_VERSION = "0.153.0";
+var CODEX_PROTOCOL_HEADERS = {
+  "OpenAI-Beta": "responses=experimental",
+  originator: "0",
+  version: CODEX_CLIENT_VERSION
+};
 var CODEX_OAUTH_ISSUER = "https://auth.openai.com";
 var CODEX_OAUTH_CLIENT_ID = "app_EMoamEEZ73f0CkXaXp7hrann";
 var CODEX_DEFAULT_MODEL = "gpt-5.5";
 var LOOP_SERVER_COMPACTION_TOKENS = 15e4;
-var PROCESS_SESSION_ID = `0sec-${Math.random().toString(36).slice(2, 10)}-${Date.now().toString(36)}`;
+var PROCESS_SESSION_ID = `0-${Math.random().toString(36).slice(2, 10)}-${Date.now().toString(36)}`;
 var chatGptCodexAuthStates = /* @__PURE__ */ new Map();
 function codexAuthStateKey(state) {
   return JSON.stringify([state.authFilePath, state.accountId, state.refreshToken || state.accessToken]);
 }
 function readChatGptCodexEnv(env2 = process.env) {
-  const access = env2["0SEC_CHATGPT_ACCESS_TOKEN"];
-  const refresh = env2["0SEC_CHATGPT_OAUTH_REFRESH_TOKEN"];
+  const access = env2["ZERO_CHATGPT_ACCESS_TOKEN"];
+  const refresh = env2["ZERO_CHATGPT_OAUTH_REFRESH_TOKEN"];
   if ((!access || access.length === 0) && (!refresh || refresh.length === 0)) {
     return void 0;
   }
-  const accountId = env2["0SEC_CHATGPT_ACCOUNT_ID"];
+  const accountId = env2["ZERO_CHATGPT_ACCOUNT_ID"];
   return {
     accessToken: access && access.length > 0 ? access : void 0,
     refreshToken: refresh && refresh.length > 0 ? refresh : void 0,
@@ -2054,13 +1715,13 @@ function readChatGptCodexEnv(env2 = process.env) {
   };
 }
 function resolveChatGptCodexAuthPath(env2 = process.env) {
-  return env2["0SEC_CHATGPT_AUTH_FILE"] ?? join2(env2.HOME ?? homedir(), ".codex", "auth.json");
+  return env2["ZERO_CHATGPT_AUTH_FILE"] ?? join(env2.HOME ?? homedir(), ".codex", "auth.json");
 }
 function persistChatGptCodexAuthFile(authPath, tokens, usedRefreshToken) {
   try {
     let existing = {};
     try {
-      existing = JSON.parse(readFileSync2(authPath, "utf8"));
+      existing = JSON.parse(readFileSync(authPath, "utf8"));
     } catch {
       existing = {};
     }
@@ -2083,7 +1744,7 @@ function persistChatGptCodexAuthFile(authPath, tokens, usedRefreshToken) {
 `, { mode: 384 });
     renameSync(tmp, authPath);
   } catch (err) {
-    process.stderr.write(`[0sec] warning: could not persist rotated Codex refresh token to ${authPath}: ${err instanceof Error ? err.message : String(err)}
+    process.stderr.write(`[0] warning: could not persist rotated Codex refresh token to ${authPath}: ${err instanceof Error ? err.message : String(err)}
 `);
   }
 }
@@ -2092,7 +1753,7 @@ function readChatGptCodexAuthFile(env2 = process.env) {
   if (!existsSync(authPath))
     return void 0;
   try {
-    const auth = JSON.parse(readFileSync2(authPath, "utf8"));
+    const auth = JSON.parse(readFileSync(authPath, "utf8"));
     const tokens = auth.tokens;
     if (!tokens)
       return void 0;
@@ -2119,6 +1780,7 @@ function accessTokenExpiryMs(accessToken) {
 }
 async function refreshChatGptCodexAccessToken(refreshToken) {
   const res = await fetch(`${CODEX_OAUTH_ISSUER}/oauth/token`, {
+    signal: AbortSignal.timeout(3e4),
     method: "POST",
     headers: { "Content-Type": "application/x-www-form-urlencoded" },
     body: new URLSearchParams({
@@ -2170,12 +1832,15 @@ function extractChatGptAccountId(tokens) {
   }
   return void 0;
 }
+async function getChatGptCodexAccessToken(env2 = process.env) {
+  return refreshChatGptCodexAuthState(resolveChatGptCodexAuthState(env2));
+}
 function resolveChatGptCodexAuthState(env2) {
   const fromEnvOnly = readChatGptCodexEnv(env2);
   const fromFile = fromEnvOnly ? void 0 : readChatGptCodexAuthFile(env2);
   const tokens = fromEnvOnly ?? fromFile;
   if (!tokens) {
-    throw new Error("ChatGPT Codex auth: neither 0SEC_CHATGPT_ACCESS_TOKEN nor 0SEC_CHATGPT_OAUTH_REFRESH_TOKEN is set. Run `codex login` and either forward the access token via worker-controller (preferred for multi-sandbox dispatch \u2014 avoids the OAuth refresh-token rotation race) or keep a valid ~/.codex/auth.json on this host.");
+    throw new Error("ChatGPT Codex auth: neither ZERO_CHATGPT_ACCESS_TOKEN nor ZERO_CHATGPT_OAUTH_REFRESH_TOKEN is set. Run `codex login` and either forward the access token via worker-controller (preferred for multi-sandbox dispatch \u2014 avoids the OAuth refresh-token rotation race) or keep a valid ~/.codex/auth.json on this host.");
   }
   const identity = {
     refreshToken: tokens.refreshToken ?? "",
@@ -2235,8 +1900,8 @@ function geminiAuthStateKey(refreshToken, accessToken) {
   return JSON.stringify([refreshToken, refreshToken ? void 0 : accessToken]);
 }
 function readGeminiCodeAssistEnv(env2 = process.env) {
-  const access = env2["0SEC_GEMINI_ACCESS_TOKEN"];
-  const refresh = env2["0SEC_GEMINI_OAUTH_REFRESH_TOKEN"];
+  const access = env2["ZERO_GEMINI_ACCESS_TOKEN"];
+  const refresh = env2["ZERO_GEMINI_OAUTH_REFRESH_TOKEN"];
   if ((!access || access.length === 0) && (!refresh || refresh.length === 0))
     return void 0;
   return {
@@ -2247,7 +1912,7 @@ function readGeminiCodeAssistEnv(env2 = process.env) {
 function resolveGeminiCodeAssistAuthState(env2) {
   const tokens = readGeminiCodeAssistEnv(env2);
   if (!tokens) {
-    throw new Error("Google Gemini Code Assist auth: neither 0SEC_GEMINI_ACCESS_TOKEN nor 0SEC_GEMINI_OAUTH_REFRESH_TOKEN is set. Sign in with your Google account (0sec connect) or forward a fresh access token.");
+    throw new Error("Google Gemini Code Assist auth: neither ZERO_GEMINI_ACCESS_TOKEN nor ZERO_GEMINI_OAUTH_REFRESH_TOKEN is set. Sign in with your Google account (0 connect) or forward a fresh access token.");
   }
   const key = geminiAuthStateKey(tokens.refreshToken ?? "", tokens.accessToken);
   const existing = geminiCodeAssistAuthStates.get(key);
@@ -2360,7 +2025,7 @@ async function resolveGeminiCodeAssistProject(state, env2, sleep = (ms) => new P
     return state.inflightProjectResolve;
   state.inflightProjectResolve = (async () => {
     try {
-      const override = firstNonEmptyEnv(env2, "GOOGLE_CLOUD_PROJECT", "0SEC_GEMINI_PROJECT");
+      const override = firstNonEmptyEnv(env2, "GOOGLE_CLOUD_PROJECT", "ZERO_GEMINI_PROJECT");
       const accessToken = await refreshGeminiCodeAssistAuthState(state);
       let load;
       try {
@@ -2372,7 +2037,7 @@ async function resolveGeminiCodeAssistProject(state, env2, sleep = (ms) => new P
         if (err.securityPolicyViolated) {
           if (override)
             return override;
-          throw new Error("Google Gemini Code Assist: this account is behind a VPC Service Controls perimeter \u2014 set GOOGLE_CLOUD_PROJECT (or 0SEC_GEMINI_PROJECT).");
+          throw new Error("Google Gemini Code Assist: this account is behind a VPC Service Controls perimeter \u2014 set GOOGLE_CLOUD_PROJECT (or ZERO_GEMINI_PROJECT).");
         }
         throw err;
       }
@@ -2434,7 +2099,7 @@ function parseCodexAzureConfig(env2 = process.env) {
   if (!existsSync(configPath))
     return {};
   try {
-    const content = readFileSync2(configPath, "utf8");
+    const content = readFileSync(configPath, "utf8");
     const azureSectionMatch = content.match(/\[model_providers\.azure\]([\s\S]*?)(?:\n\[|$)/);
     const activeProviderMatch = content.match(/^\s*model_provider\s*=\s*"([^"]+)"/m);
     const baseUrlMatch = azureSectionMatch?.[1]?.match(/base_url\s*=\s*"([^"]+)"/);
@@ -2484,13 +2149,13 @@ function providerForModel(model, env2) {
     return env2.OPENCODE_API_KEY ? "opencode" : void 0;
   }
   if (m.startsWith("copilot/")) {
-    return env2["0SEC_COPILOT_GITHUB_TOKEN"] ? "copilot" : void 0;
+    return env2["ZERO_COPILOT_GITHUB_TOKEN"] ? "copilot" : void 0;
   }
   if (m.startsWith("gemini") || m.startsWith("google/")) {
-    return env2["0SEC_GEMINI_ACCESS_TOKEN"] || env2["0SEC_GEMINI_OAUTH_REFRESH_TOKEN"] ? "google" : void 0;
+    return env2["ZERO_GEMINI_ACCESS_TOKEN"] || env2["ZERO_GEMINI_OAUTH_REFRESH_TOKEN"] ? "google" : void 0;
   }
   if (/^gpt-|^o[1-4](?:[-_]|$)/.test(m)) {
-    if (env2["0SEC_CHATGPT_ACCESS_TOKEN"] || env2["0SEC_CHATGPT_OAUTH_REFRESH_TOKEN"])
+    if (readChatGptCodexEnv(env2) || readChatGptCodexAuthFile(env2))
       return "chatgpt-codex";
     if (env2.OPENAI_API_KEY)
       return "openai";
@@ -2518,33 +2183,32 @@ var DEFAULT_PROVIDER_MODELS = {
   xai: XAI_DEFAULT_MODEL,
   opencode: OPENCODE_DEFAULT_MODEL,
   copilot: COPILOT_DEFAULT_MODEL,
-  google: GEMINI_DEFAULT_MODEL,
-  hosted: ""
+  google: GEMINI_DEFAULT_MODEL
 };
 var AUTO_MODEL_SENTINEL = "auto";
 function detectProvider(configApiKey, preferredModel, env2, configProvider) {
   if (configProvider !== void 0 && !Object.hasOwn(DEFAULT_PROVIDER_MODELS, configProvider)) {
     throw new Error(`RuntimeConfig.provider is unsupported: ${configProvider}`);
   }
-  const selectedProviderRaw = configProvider ?? env2["0SEC_SELECTED_PROVIDER"]?.trim();
-  const forcedProviderRaw = env2["0SEC_FORCE_PROVIDER"]?.trim() || void 0;
+  const selectedProviderRaw = configProvider ?? env2["ZERO_SELECTED_PROVIDER"]?.trim();
+  const forcedProviderRaw = env2["ZERO_FORCE_PROVIDER"]?.trim() || void 0;
   if (selectedProviderRaw && forcedProviderRaw && selectedProviderRaw !== forcedProviderRaw) {
-    throw new Error(`${configProvider !== void 0 ? "RuntimeConfig.provider" : "0SEC_SELECTED_PROVIDER"} conflicts with 0SEC_FORCE_PROVIDER`);
+    throw new Error(`${configProvider !== void 0 ? "RuntimeConfig.provider" : "ZERO_SELECTED_PROVIDER"} conflicts with ZERO_FORCE_PROVIDER`);
   }
-  const primaryModel = env2["0SEC_MODEL"]?.trim();
+  const primaryModel = env2["ZERO_MODEL"]?.trim();
   const selectedProviderApplies = configProvider !== void 0 || !preferredModel || !primaryModel || preferredModel === primaryModel;
   const pinnedProviderRaw = forcedProviderRaw ?? (selectedProviderApplies ? selectedProviderRaw : void 0);
   if (pinnedProviderRaw) {
-    const source = pinnedProviderRaw === forcedProviderRaw ? "0SEC_FORCE_PROVIDER" : configProvider !== void 0 ? "RuntimeConfig.provider" : "0SEC_SELECTED_PROVIDER";
+    const source = pinnedProviderRaw === forcedProviderRaw ? "ZERO_FORCE_PROVIDER" : configProvider !== void 0 ? "RuntimeConfig.provider" : "ZERO_SELECTED_PROVIDER";
     if (!Object.hasOwn(DEFAULT_PROVIDER_MODELS, pinnedProviderRaw)) {
       throw new Error(`${source} is unsupported: ${pinnedProviderRaw}`);
     }
     const provider = pinnedProviderRaw;
-    const model = preferredModel ?? env2["0SEC_MODEL"] ?? (configProvider !== void 0 || provider === "hosted" ? DEFAULT_PROVIDER_MODELS[provider] : void 0);
-    if (model === void 0 || model === "" && provider !== "hosted") {
+    const model = preferredModel ?? env2["ZERO_MODEL"] ?? (configProvider !== void 0 ? DEFAULT_PROVIDER_MODELS[provider] : void 0);
+    if (model === void 0 || model === "") {
       throw new Error(`${source} requires an explicit model`);
     }
-    if (configApiKey && (provider === "hosted" || provider === "chatgpt-codex")) {
+    if (configApiKey && provider === "chatgpt-codex") {
       throw new Error(`${source}=${provider} requires its own authentication, not RuntimeConfig.apiKey`);
     }
     const resolved = resolveFailoverProvider(provider, model, env2, configApiKey);
@@ -2650,7 +2314,7 @@ function detectProvider(configApiKey, preferredModel, env2, configProvider) {
     case "copilot":
       return {
         provider: "copilot",
-        apiKey: env2["0SEC_COPILOT_GITHUB_TOKEN"],
+        apiKey: env2["ZERO_COPILOT_GITHUB_TOKEN"],
         baseUrl: env2.COPILOT_BASE_URL ?? COPILOT_API_BASE,
         defaultModel: preferredModel ?? COPILOT_DEFAULT_MODEL,
         wireApi: "chat_completions"
@@ -2668,7 +2332,7 @@ function detectProvider(configApiKey, preferredModel, env2, configProvider) {
         provider: "chatgpt-codex",
         apiKey: "",
         baseUrl: CODEX_API_ENDPOINT,
-        defaultModel: env2["0SEC_MODEL"] ?? CODEX_DEFAULT_MODEL,
+        defaultModel: env2["ZERO_MODEL"] ?? CODEX_DEFAULT_MODEL,
         wireApi: "responses"
       };
     case "anthropic":
@@ -2698,8 +2362,8 @@ function detectProvider(configApiKey, preferredModel, env2, configProvider) {
     default:
       break;
   }
-  const chatGptAccess = env2["0SEC_CHATGPT_ACCESS_TOKEN"];
-  const chatGptRefresh = env2["0SEC_CHATGPT_OAUTH_REFRESH_TOKEN"];
+  const chatGptAccess = env2["ZERO_CHATGPT_ACCESS_TOKEN"];
+  const chatGptRefresh = env2["ZERO_CHATGPT_OAUTH_REFRESH_TOKEN"];
   const chatGptAuthFile = !chatGptAccess && !chatGptRefresh ? readChatGptCodexAuthFile(env2) : void 0;
   if (chatGptAccess && chatGptAccess.length > 0 || chatGptRefresh && chatGptRefresh.length > 0 || !!chatGptAuthFile) {
     return {
@@ -2712,7 +2376,7 @@ function detectProvider(configApiKey, preferredModel, env2, configProvider) {
       // baseUrl is informational only — the runtime hardcodes
       // CODEX_API_ENDPOINT for this provider.
       baseUrl: CODEX_API_ENDPOINT,
-      defaultModel: env2["0SEC_MODEL"] ?? CODEX_DEFAULT_MODEL,
+      defaultModel: env2["ZERO_MODEL"] ?? CODEX_DEFAULT_MODEL,
       wireApi: "responses"
     };
   }
@@ -2808,7 +2472,7 @@ function detectProvider(configApiKey, preferredModel, env2, configProvider) {
       wireApi: opencodeWireApiForModel(preferredModel)
     };
   }
-  const copilotToken = env2["0SEC_COPILOT_GITHUB_TOKEN"];
+  const copilotToken = env2["ZERO_COPILOT_GITHUB_TOKEN"];
   if (copilotToken) {
     return {
       provider: "copilot",
@@ -2818,14 +2482,14 @@ function detectProvider(configApiKey, preferredModel, env2, configProvider) {
       wireApi: "chat_completions"
     };
   }
-  const geminiAccess = env2["0SEC_GEMINI_ACCESS_TOKEN"];
-  const geminiRefresh = env2["0SEC_GEMINI_OAUTH_REFRESH_TOKEN"];
+  const geminiAccess = env2["ZERO_GEMINI_ACCESS_TOKEN"];
+  const geminiRefresh = env2["ZERO_GEMINI_OAUTH_REFRESH_TOKEN"];
   if (geminiAccess && geminiAccess.length > 0 || geminiRefresh && geminiRefresh.length > 0) {
     return {
       provider: "google",
       apiKey: "",
       baseUrl: CODE_ASSIST_ENDPOINT,
-      defaultModel: env2["0SEC_MODEL"] ?? GEMINI_DEFAULT_MODEL,
+      defaultModel: env2["ZERO_MODEL"] ?? GEMINI_DEFAULT_MODEL,
       wireApi: "google_generate_content"
     };
   }
@@ -2839,21 +2503,6 @@ function detectProvider(configApiKey, preferredModel, env2, configProvider) {
       wireApi: "chat_completions"
     };
   }
-  try {
-    const hostedCreds = loadCloudCredentials({
-      env: env2,
-      warn: () => {
-      }
-    });
-    return {
-      provider: "hosted",
-      apiKey: hostedCreds.token,
-      baseUrl: `${hostedCreds.host}/api/inference/v1`,
-      defaultModel: "",
-      wireApi: "chat_completions"
-    };
-  } catch {
-  }
   return {
     provider: "anthropic",
     apiKey: "",
@@ -2876,27 +2525,23 @@ var LlmApiRuntime = class _LlmApiRuntime {
   apiKey;
   baseUrl;
   model;
+  pricingModelSource = "";
+  pricingProviderSource;
+  pricingModelKey = "";
   wireApi;
   reasoningEffort;
   azureConfig;
   serverCompactionTokens;
-  /** Ordered fallback chain (0SEC_LLM_FALLBACK). Empty = no failover. */
+  /** Ordered fallback chain (ZERO_LLM_FALLBACK). Empty = no failover. */
   fallbackChain;
   /** Index into fallbackChain — which entry to try next. */
   fallbackIndex;
-  /** Resolve and validate the hosted model and wire protocol once per runtime. */
-  hostedCatalogPromise = null;
-  /** Catalog ceiling, resolved before hosted inference is submitted. */
-  hostedMaxOutputTokens;
   constructor(config, inherited) {
     if (inherited) {
       const timeout = config.timeout ?? inherited.config.timeout ?? 12e4;
       if (!Number.isFinite(timeout) || timeout <= 0) {
         throw new Error("Subagent timeout must be a positive finite number");
       }
-      if (inherited.provider === "hosted" && inherited.hostedMaxOutputTokens === void 0) {
-        throw new Error("Hosted model catalog must resolve before creating a subagent");
-      }
       const model = config.model ?? inherited.model;
       const modelChanged = model !== inherited.model;
       this.config = {
@@ -2930,10 +2575,6 @@ var LlmApiRuntime = class _LlmApiRuntime {
         this.geminiAuthState = inherited.geminiAuthState;
       this.fallbackChain = [];
       this.fallbackIndex = 0;
-      if (!modelChanged) {
-        this.hostedMaxOutputTokens = inherited.hostedMaxOutputTokens;
-        this.hostedCatalogPromise = inherited.hostedCatalogPromise;
-      }
       return;
     }
     this.applyConfiguration(config);
@@ -2956,7 +2597,7 @@ var LlmApiRuntime = class _LlmApiRuntime {
       credentials: resolveFailoverProvider(entry.provider, entry.model, this.env)
     }));
     this.fallbackIndex = 0;
-    const detected = detectProvider(config.apiKey, config.model ?? this.env["0SEC_MODEL"], this.env, config.provider);
+    const detected = detectProvider(config.apiKey, config.model ?? this.env["ZERO_MODEL"], this.env, config.provider);
     this.provider = detected.provider;
     this.apiKey = detected.apiKey;
     this.baseUrl = detected.baseUrl;
@@ -2973,9 +2614,9 @@ var LlmApiRuntime = class _LlmApiRuntime {
         this.geminiAuthState = resolveGeminiCodeAssistAuthState(this.env);
       }
     }
-    this.reasoningEffort = this.env["0SEC_REASONING_EFFORT"] ?? detected.reasoningEffort;
+    this.reasoningEffort = this.env["ZERO_REASONING_EFFORT"] ?? detected.reasoningEffort;
     this.serverCompactionTokens = config.serverCompactionTokens !== void 0 ? Math.max(1e3, config.serverCompactionTokens) : void 0;
-    const requestedModel = config.model ?? this.env["0SEC_MODEL"];
+    const requestedModel = config.model ?? this.env["ZERO_MODEL"];
     if (requestedModel === "free" && this.provider === "openrouter") {
       this.model = FREE_OPENROUTER_MODEL;
     } else {
@@ -2988,16 +2629,51 @@ var LlmApiRuntime = class _LlmApiRuntime {
       this.model = copilotModelId(this.model);
     }
     this.applyModelWireApi();
-    if (this.apiKey && !this.env["0SEC_SKIP_PROVIDER_BANNER"]) {
+    if (this.apiKey && !this.env["ZERO_SKIP_PROVIDER_BANNER"]) {
       void logProviderStartup(this.provider, this.providerLabel, this.baseUrl, this.model, this.wireApi, this.apiKey).catch(() => {
       });
     }
   }
+  /** Non-secret identity for preferences scoped to the captured connection. */
+  connectionIdentity() {
+    let identity;
+    if (this.provider === "chatgpt-codex") {
+      const state = this.codexAuthState;
+      if (!state)
+        return void 0;
+      if (state.accountId) {
+        identity = [this.provider, "account", state.accountId];
+      } else {
+        const credential = state.refreshToken || state.accessToken;
+        if (!credential)
+          return void 0;
+        identity = [this.provider, "credential", credential];
+      }
+    } else if (this.provider === "google") {
+      const credential = this.geminiAuthState?.refreshToken || this.geminiAuthState?.accessToken;
+      if (!credential)
+        return void 0;
+      identity = [this.provider, this.baseUrl, credential, firstNonEmptyEnv(this.env, "GOOGLE_CLOUD_PROJECT", "ZERO_GEMINI_PROJECT") ?? ""];
+    } else {
+      if (!this.apiKey)
+        return void 0;
+      identity = [this.provider, this.baseUrl, this.apiKey];
+    }
+    return createHash("sha256").update(JSON.stringify(identity)).digest("hex");
+  }
+  /** Discover models using this runtime's captured account, including after a separate login changes. */
+  async codexModelCatalog(signal) {
+    const state = this.codexAuthState;
+    if (this.provider !== "chatgpt-codex" || !state)
+      throw new Error("No active Codex subscription");
+    const { loadCodexModelCatalog } = await import("./codex-models-7TLJSY3L.js");
+    return loadCodexModelCatalog({ signal, resolveCredentials: () => refreshChatGptCodexAuthState(state) });
+  }
   /**
    * Mutate the live selection in place so the NEXT turn (the engine reads
    * `config.runtime` per turn) and the NEXT `forkForSubagent` pick up the new
-   * model / provider / role map with zero session teardown. No-op-safe:
-   * undefined fields leave the corresponding state unchanged.
+   * model / provider / role map with zero session teardown. Undefined fields
+   * preserve the current state, except a provider change resets an omitted model.
    */
   reconfigure(sel) {
     const providerChanged = sel.provider !== void 0 && sel.provider !== this.provider;
@@ -3005,14 +2681,12 @@ var LlmApiRuntime = class _LlmApiRuntime {
       const merged = {
         ...this.config,
         apiKey: void 0,
-        provider: sel.provider,
-        ...sel.model !== void 0 ? { model: sel.model } : {},
+        provider: sel.provider ?? this.provider,
+        model: sel.model,
         ...sel.agentModels !== void 0 ? { agentModels: sel.agentModels } : {},
         ...sel.singleModel !== void 0 ? { singleModel: sel.singleModel } : {},
         ...sel.env !== void 0 ? { env: sel.env } : {}
       };
-      this.hostedCatalogPromise = null;
-      this.hostedMaxOutputTokens = void 0;
       this.applyConfiguration(merged);
       return;
     }
@@ -3038,10 +2712,10 @@ var LlmApiRuntime = class _LlmApiRuntime {
       }
     }
   }
-  /** Exact deployments requiring Responses for tools, shared by roots and forks. */
+  /** Modern OpenAI tool calls use Responses, shared by roots and forks. */
   applyModelWireApi() {
     const normalizedModel = this.model.toLowerCase();
-    if (this.wireApi === "chat_completions" && (this.provider === "azure" && normalizedModel === "gpt-5.6-sol" || this.provider === "openai" && normalizedModel === "gpt-5.6-luna")) {
+    if (this.wireApi === "chat_completions" && (this.provider === "azure" && normalizedModel === "gpt-5.6-sol" || this.provider === "openai" && (normalizedModel === "gpt-5.6-luna" || /^gpt-(?:[6-9]|\d{2,})(?:[-.]|$)/.test(normalizedModel)))) {
       this.wireApi = "responses";
     }
   }
@@ -3050,18 +2724,14 @@ var LlmApiRuntime = class _LlmApiRuntime {
    * Providers whose credentials are present in THIS runtime's environment
    * (`this.env`, never process-global). A provider counts as accessible iff
    * `resolveFailoverProvider` — the same auth-presence check the cross-provider
-   * failover chain uses — can build a connection for it, so auth-only providers
-   * (chatgpt-codex OAuth) and cloud-hosted are covered by the identical rule.
+   * failover chain uses — including auth-only providers (chatgpt-codex OAuth).
    */
   accessibleProviders() {
     const out = [];
     for (const provider of Object.keys(DEFAULT_PROVIDER_MODELS)) {
       const probeModel = DEFAULT_PROVIDER_MODELS[provider] || "probe";
-      try {
-        if (resolveFailoverProvider(provider, probeModel, this.env))
-          out.push(provider);
-      } catch {
-      }
+      if (resolveFailoverProvider(provider, probeModel, this.env))
+        out.push(provider);
     }
     return out;
   }
@@ -3098,7 +2768,6 @@ var LlmApiRuntime = class _LlmApiRuntime {
     if (!Number.isFinite(timeoutMs) || timeoutMs <= 0) {
       throw new Error("Subagent timeout must be a positive finite number");
     }
-    await this.ensureHostedModel();
     const roleModel = selection?.role !== void 0 && this.config.agentModels && Object.hasOwn(this.config.agentModels, selection.role) ? this.config.agentModels[selection.role] : void 0;
     const autoInEffect = roleModel === AUTO_MODEL_SENTINEL || this.config.autoRoute === true && roleModel === void 0;
     const configuredModel = roleModel === AUTO_MODEL_SENTINEL ? void 0 : roleModel;
@@ -3114,45 +2783,15 @@ var LlmApiRuntime = class _LlmApiRuntime {
       throw new Error(autoInEffect ? `Subagent model "${selectedModel}" is not reachable: its provider has no configured credentials.` : `Subagent model "${selectedModel}" is not operator-approved. Configure agentModels before selecting it.`);
     }
     const child = new _LlmApiRuntime({ type: "api", timeout: timeoutMs, model }, this);
-    await child.ensureHostedModel();
     return child;
   }
-  /** The server catalog is authoritative even when a model was selected explicitly. */
-  async ensureHostedModel() {
-    if (this.provider !== "hosted")
-      return;
-    if (!this.hostedCatalogPromise) {
-      this.hostedCatalogPromise = (async () => {
-        const client = new CloudClient({
-          host: this.baseUrl.replace(/\/api\/inference\/v1$/, ""),
-          token: this.apiKey
-        });
-        const catalog = await client.getInferenceModels();
-        const selected = this.model ? catalog.data.find((model) => model.id === this.model) : catalog.data[0];
-        if (!selected) {
-          throw new Error(this.model ? `Hosted model "${this.model}" is unavailable. Run \`0sec models\` for available models.` : "No hosted models are available. Run `0sec models` to check service availability.");
-        }
-        this.model = selected.id;
-        this.wireApi = selected.wire_api;
-        this.hostedMaxOutputTokens = selected.max_output_tokens;
-      })();
-    }
-    await this.hostedCatalogPromise;
-  }
   /**
    * A hard dollar ceiling needs a provider-enforced bound on the next response.
    * ChatGPT Codex OAuth rejects `max_output_tokens`, so it cannot support that
    * contract; callers must fail closed before making a metered comparison call.
    */
   get outputTokenLimit() {
-    return this.provider === "chatgpt-codex" ? void 0 : this.effectiveOutputTokens;
-  }
-  /** Hosted requests honor both the catalog ceiling and the local hard cap. */
-  get effectiveOutputTokens() {
-    if (this.provider === "hosted" && this.hostedMaxOutputTokens !== void 0) {
-      return Math.min(NATIVE_COMPLETION_TOKEN_LIMIT, this.hostedMaxOutputTokens);
-    }
-    return NATIVE_COMPLETION_TOKEN_LIMIT;
+    return this.provider === "chatgpt-codex" ? void 0 : NATIVE_COMPLETION_TOKEN_LIMIT;
   }
   /**
    * Whether this provider uses OpenAI-compatible chat/completions format.
@@ -3164,7 +2803,7 @@ var LlmApiRuntime = class _LlmApiRuntime {
    * inert default — it is intentionally unused for these two providers.
    */
   get isOpenAICompat() {
-    return this.provider === "openrouter" || this.provider === "openai" || this.provider === "azure" || this.provider === "deepseek" || this.provider === "qwen" || this.provider === "xai" || this.provider === "copilot" || this.provider === "hosted" || this.provider === "opencode" && (this.wireApi === "chat_completions" || this.wireApi === "responses") || // chatgpt-codex always speaks Responses API; treat it as
+    return this.provider === "openrouter" || this.provider === "openai" || this.provider === "azure" || this.provider === "deepseek" || this.provider === "qwen" || this.provider === "xai" || this.provider === "copilot" || this.provider === "opencode" && (this.wireApi === "chat_completions" || this.wireApi === "responses") || // chatgpt-codex always speaks Responses API; treat it as
     // OpenAI-compat for body-shape branching purposes (the Responses
     // wire-API code paths below already key on `wireApi === "responses"`
     // and produce a body codex's backend accepts as-is).
@@ -3210,13 +2849,25 @@ var LlmApiRuntime = class _LlmApiRuntime {
   resolvedModel() {
     return this.model;
   }
+  /** Price discovered IDs on their actual route without changing the wire model. */
+  resolvedPricingModel() {
+    if (this.provider === "chatgpt-codex" || this.provider === "copilot" || this.provider === "google")
+      return this.model;
+    if (this.pricingModelSource !== this.model || this.pricingProviderSource !== this.provider) {
+      this.pricingModelSource = this.model;
+      this.pricingProviderSource = this.provider;
+      this.pricingModelKey = `${this.provider}/${this.model}`;
+    }
+    return this.pricingModelKey;
+  }
   /** Build the appropriate headers for the configured provider. */
   buildHeaders() {
     if (this.provider === "chatgpt-codex") {
       return {
         "Content-Type": "application/json",
-        originator: "0sec",
-        "User-Agent": `0sec/${VERSION}`
+        ...CODEX_PROTOCOL_HEADERS,
+        "x-codex-routing-hint": `model=${this.model}`,
+        "User-Agent": `0/${VERSION}`
       };
     }
     if (this.isGeminiCodeAssist) {
@@ -3248,8 +2899,8 @@ var LlmApiRuntime = class _LlmApiRuntime {
         headers["Authorization"] = `Bearer ${this.apiKey}`;
       }
       if (this.provider === "openrouter") {
-        headers["HTTP-Referer"] = "https://0sec.ai";
-        headers["X-Title"] = "0sec Security Scanner";
+        headers["HTTP-Referer"] = "https://0.security";
+        headers["X-Title"] = "0 Security Scanner";
       }
       return headers;
     }
@@ -3273,7 +2924,7 @@ var LlmApiRuntime = class _LlmApiRuntime {
    * happens.
    *
    * session_id is process-stable (PROCESS_SESSION_ID, randomised
-   * once at module load). A 0sec-cli invocation = one scan = one
+   * once at module load). A @0/cli invocation = one scan = one
    * session, so the process-lifetime constant is the right
    * granularity. If we ever want per-scan ids inside a long-lived
    * controller process, add a setter on the runtime; for now this
@@ -3427,7 +3078,7 @@ var LlmApiRuntime = class _LlmApiRuntime {
   /**
    * Per-turn prompt-cache accounting line, so a run can be shown to actually
    * be hitting cache rather than assumed to be. Off unless
-   * `0SEC_DEBUG_PROMPT_CACHE` is set — this fires once per agent turn, and an
+   * `ZERO_DEBUG_PROMPT_CACHE` is set — this fires once per agent turn, and an
    * unconditional line would interleave with the TUI on every scan.
    *
    * The same numbers reach the cloud without this flag: `cachedInputTokens`
@@ -3435,7 +3086,7 @@ var LlmApiRuntime = class _LlmApiRuntime {
    * is the durable, queryable proof. This is the local fast path.
    */
   logCacheUsage(usage) {
-    if (!usage || !process.env["0SEC_DEBUG_PROMPT_CACHE"])
+    if (!usage || !process.env["ZERO_DEBUG_PROMPT_CACHE"])
       return;
     const read = usage.cachedInputTokens ?? 0;
     const write = usage.cacheWriteTokens ?? 0;
@@ -3478,12 +3129,10 @@ var LlmApiRuntime = class _LlmApiRuntime {
         return "GitHub Copilot";
       case "google":
         return "Google Gemini (Code Assist)";
-      case "hosted":
-        return "0sec Cloud";
     }
   }
   noKeyError() {
-    return "No provider credential found. Set one of:\n  env 0SEC_CHATGPT_OAUTH_REFRESH_TOKEN=... 0sec <command> (ChatGPT Codex subscription auth)\n  export OPENROUTER_API_KEY=sk-or-...   (OpenRouter \u2014 many models, one key)\n  export DEEPSEEK_API_KEY=...           (DeepSeek \u2014 direct Flash 0731 inference)\n  export ANTHROPIC_API_KEY=sk-ant-...    (Anthropic \u2014 direct Claude access)\n  export AZURE_OPENAI_API_KEY=...        (Azure OpenAI \u2014 reuse your Codex Azure provider)\n  export OPENAI_API_KEY=sk-...           (OpenAI \u2014 direct GPT access)\n  export Z_AI_API_KEY=...                (Z.ai GLM \u2014 flat-rate Coding Plan, Anthropic-compatible)\n  export KIMI_API_KEY=...                (Moonshot Kimi K3 \u2014 flat-rate coding, Anthropic-compatible)\n  export QWEN_API_KEY=...                (Alibaba Qwen \u2014 Token Plan sub, OpenAI-compatible)\n  export XAI_API_KEY=...                 (xAI Grok \u2014 OpenAI-compatible)\n  export OPENCODE_API_KEY=...            (OpenCode Zen \u2014 multi-wire gateway)\n  export 0SEC_COPILOT_GITHUB_TOKEN=...   (GitHub Copilot \u2014 device-code OAuth token)\n  Run `0sec login`                     (0sec hosted inference)";
+    return "No provider credential found. Set one of:\n  env ZERO_CHATGPT_OAUTH_REFRESH_TOKEN=... 0 <command> (ChatGPT Codex subscription auth)\n  export OPENROUTER_API_KEY=sk-or-...   (OpenRouter \u2014 many models, one key)\n  export DEEPSEEK_API_KEY=...           (DeepSeek \u2014 direct Flash 0731 inference)\n  export ANTHROPIC_API_KEY=sk-ant-...    (Anthropic \u2014 direct Claude access)\n  export AZURE_OPENAI_API_KEY=...        (Azure OpenAI \u2014 reuse your Codex Azure provider)\n  export OPENAI_API_KEY=sk-...           (OpenAI \u2014 direct GPT access)\n  export Z_AI_API_KEY=...                (Z.ai GLM \u2014 flat-rate Coding Plan, Anthropic-compatible)\n  export KIMI_API_KEY=...                (Moonshot Kimi K3 \u2014 flat-rate coding, Anthropic-compatible)\n  export QWEN_API_KEY=...                (Alibaba Qwen \u2014 Token Plan sub, OpenAI-compatible)\n  export XAI_API_KEY=...                 (xAI Grok \u2014 OpenAI-compatible)\n  export OPENCODE_API_KEY=...            (OpenCode Zen \u2014 multi-wire gateway)\n  export ZERO_COPILOT_GITHUB_TOKEN=...   (GitHub Copilot \u2014 device-code OAuth token)";
   }
   getConfigurationDiagnostics() {
     if (!this.apiKey && this.provider !== "chatgpt-codex") {
@@ -3503,7 +3152,7 @@ var LlmApiRuntime = class _LlmApiRuntime {
       };
     }
     const hasConfiguredBaseUrl = !!(this.env.AZURE_OPENAI_BASE_URL || this.env.OPENAI_BASE_URL || this.azureConfig.baseUrl);
-    const hasConfiguredModel = !!(this.config.model || this.env["0SEC_MODEL"] || this.env.AZURE_OPENAI_MODEL || this.azureConfig.model);
+    const hasConfiguredModel = !!(this.config.model || this.env["ZERO_MODEL"] || this.env.AZURE_OPENAI_MODEL || this.azureConfig.model);
     const missing = [];
     if (!hasConfiguredBaseUrl) {
       missing.push("AZURE_OPENAI_BASE_URL (or [model_providers.azure].base_url in ~/.codex/config.toml)");
@@ -3519,7 +3168,7 @@ var LlmApiRuntime = class _LlmApiRuntime {
         reason: "invalid_config",
         fatalError: `Azure OpenAI runtime is selected, but the configuration is incomplete.
 Missing: ${missing.join("; ")}
-0sec will not guess Azure defaults because that can silently route to the wrong endpoint or deployment.`
+0 will not guess Azure defaults because that can silently route to the wrong endpoint or deployment.`
       };
     }
     return {
@@ -3537,15 +3186,15 @@ Missing: ${missing.join("; ")}
    *
    * Two 429 classes are handled differently:
    * - per-minute rate limit → retry with the wider 429 budget
-   *   (0SEC_LLM_429_MAX_RETRIES attempts / 0SEC_LLM_429_MAX_RETRY_WAIT_MS
+   *   (ZERO_LLM_429_MAX_RETRIES attempts / ZERO_LLM_429_MAX_RETRY_WAIT_MS
    *   cumulative, defaults 12 / 5min) since the limiter resets every ~60s;
    *   `Retry-After` / `retry-after-ms` headers are honored up to a 120s cap.
    * - plan-quota exhaustion (`usage_limit_reached`, resets in hours/days) →
-   *   skips retries and immediately advances `0SEC_LLM_FALLBACK`; if no
+   *   skips retries and immediately advances `ZERO_LLM_FALLBACK`; if no
    *   configured fallback has credentials, it throws QuotaExhaustedError.
    *
    * Other retryable statuses (transient 5xx) keep the generic budget:
-   * 0SEC_LLM_MAX_RETRIES (attempts) and 0SEC_LLM_MAX_RETRY_WAIT_MS
+   * ZERO_LLM_MAX_RETRIES (attempts) and ZERO_LLM_MAX_RETRY_WAIT_MS
    * (cumulative backoff). On exhaustion it returns the last still-failing
    * Response with its body intact, so the caller's existing `!res.ok` branch
    * surfaces the clear "API error <status>" message — a rate-limit never
@@ -3556,7 +3205,7 @@ Missing: ${missing.join("; ")}
    * is fixed across attempts.
    */
   /**
-   * Try the next fallback provider in the chain (0SEC_LLM_FALLBACK).
+   * Try the next fallback provider in the chain (ZERO_LLM_FALLBACK).
    * Updates `this.provider`, `this.model`, `this.apiKey`, `this.baseUrl`,
    * `this.wireApi` to match the next valid provider. Returns `true` when a
    * valid next provider was found and switched to, `false` when the chain is
@@ -3568,7 +3217,7 @@ Missing: ${missing.join("; ")}
       this.fallbackIndex++;
       const cfg = entry.credentials;
       if (!cfg) {
-        diag.warn("failover_provider_skipped", `0SEC_LLM_FALLBACK: skipping ${entry.provider} (auth env missing)`, { provider: entry.provider, model: entry.model, cause: "auth-env-missing" });
+        diag.warn("failover_provider_skipped", `ZERO_LLM_FALLBACK: skipping ${entry.provider} (auth env missing)`, { provider: entry.provider, model: entry.model, cause: "auth-env-missing" });
         continue;
       }
       this.provider = entry.provider;
@@ -3576,7 +3225,6 @@ Missing: ${missing.join("; ")}
       this.apiKey = cfg.apiKey;
       this.baseUrl = cfg.baseUrl;
       this.wireApi = cfg.wireApi;
-      this.hostedCatalogPromise = null;
       diag.warn("failover_engaged", `${reason} \u2014 failover to ${entry.provider} (${entry.model})`, { reason, provider: entry.provider, model: entry.model });
       return true;
     }
@@ -3590,9 +3238,8 @@ Missing: ${missing.join("; ")}
    * hammering the limit in lockstep).
    *
    * The body factory is valid only for the current provider and wire protocol.
-   * A null result signals failover: the caller resolves the hosted catalog and
-   * rebuilds the complete request under its existing timeout/cancellation signal.
-   *
+   * A null result signals failover so the caller rebuilds the complete request
+   * under its existing timeout/cancellation signal.
    * Retry + failover caps documented on `retryBackoffMs` / `llm429MaxRetries`.
    *
    * Headers are re-resolved per attempt (via ensureFreshHeaders → OAuth
@@ -3613,9 +3260,6 @@ Missing: ${missing.join("; ")}
         });
       } catch (error) {
         abort?.throwIfCancelled();
-        if (this.provider === "hosted") {
-          throw new Error("0sec hosted request outcome is unknown. Automatic replay is disabled; check your inference usage before retrying.", { cause: error });
-        }
         const cause = error instanceof Error ? error.cause : void 0;
         const causeCode = cause && typeof cause === "object" && "code" in cause && typeof cause.code === "string" ? cause.code : "unknown";
         const causeHost = cause && typeof cause === "object" && "hostname" in cause && typeof cause.hostname === "string" ? `@${cause.hostname}` : "";
@@ -3642,13 +3286,6 @@ Missing: ${missing.join("; ")}
       if (res.ok || !isRetryableHttpStatus(res.status)) {
         return res;
       }
-      if (this.provider === "hosted" && res.status === 429 && res.headers.get("x-0sec-retry-safe") !== "1") {
-        return res;
-      }
-      if (this.provider === "hosted" && res.status >= 500) {
-        await res.body?.cancel();
-        throw new Error(`0sec hosted request returned HTTP ${res.status}; its outcome may be unknown. Automatic replay is disabled; check your inference usage before retrying.`);
-      }
       abort?.throwIfCancelled();
       const is429 = res.status === 429;
       let bodyText;
@@ -3739,7 +3376,6 @@ Missing: ${missing.join("; ")}
   }
   // ── Legacy Runtime interface (single-prompt) ──
   async execute(prompt, context) {
-    await this.ensureHostedModel();
     const start = Date.now();
     if (!this.apiKey && this.provider !== "chatgpt-codex" && this.provider !== "google") {
       return {
@@ -3764,7 +3400,7 @@ Missing: ${missing.join("; ")}
           messages.push({ role: "user", content: prompt });
           res = await this.postWithRetry(() => JSON.stringify({
             model: this.model,
-            [this.maxTokensParamKey]: this.effectiveOutputTokens,
+            [this.maxTokensParamKey]: NATIVE_COMPLETION_TOKEN_LIMIT,
             messages,
             // See executeNative: explicit reasoning_effort passthrough only.
             ...this.reasoningEffort ? { reasoning_effort: this.reasoningEffort } : {}
@@ -3785,7 +3421,7 @@ Missing: ${missing.join("; ")}
           res = await this.postWithRetry(() => JSON.stringify({
             model: this.model,
             input,
-            ...isCodex ? { store: false } : { max_output_tokens: this.effectiveOutputTokens }
+            ...isCodex ? { store: false } : { max_output_tokens: NATIVE_COMPLETION_TOKEN_LIMIT }
           }), controller.signal);
         } else if (this.isGeminiCodeAssist) {
           const request = {
@@ -3812,8 +3448,6 @@ Missing: ${missing.join("; ")}
         } else {
           throw new Error(`execute: provider ${this.provider} is not mapped to a wire`);
         }
-        if (!res)
-          await this.ensureHostedModel();
       } while (!res);
       clearTimeout(timer);
       const body = await res.text();
@@ -3896,14 +3530,7 @@ Missing: ${missing.join("; ")}
       error: `${this.providerLabel} request cancelled by operator`
     };
   }
-  /**
-   * Public entry point. Runs {@link executeNativeAttempt} and, ONLY for a
-   * transient empty stream (see {@link shouldRetryNativeStream}), re-issues the
-   * whole request up to {@link llmStreamMaxAttempts} times with a short backoff.
-   * Every other outcome — success, a real API error, a timeout, an operator
-   * cancellation — is returned from the first attempt untouched, preserving the
-   * existing behaviour exactly.
-   */
+  /** Public entry point. Retries a transient empty stream with bounded backoff. */
   async executeNative(system, messages, tools, callbacks, signal) {
     const maxAttempts = llmStreamMaxAttempts();
     let result;
@@ -3924,7 +3551,6 @@ Missing: ${missing.join("; ")}
     return result;
   }
   async executeNativeAttempt(system, messages, tools, callbacks, signal) {
-    await this.ensureHostedModel();
     const start = Date.now();
     if (!this.apiKey && this.provider !== "chatgpt-codex" && this.provider !== "google") {
       return {
@@ -3989,7 +3615,7 @@ Missing: ${missing.join("; ")}
           }
           const body = {
             model: this.model,
-            [this.maxTokensParamKey]: this.effectiveOutputTokens,
+            [this.maxTokensParamKey]: NATIVE_COMPLETION_TOKEN_LIMIT,
             messages: chatMessages
           };
           if (this.reasoningEffort) {
@@ -4057,7 +3683,7 @@ Missing: ${missing.join("; ")}
           const body = {
             model: this.model,
             input,
-            ...isCodex ? { store: false, instructions: system } : { max_output_tokens: this.effectiveOutputTokens },
+            ...isCodex ? { store: false, instructions: system } : { max_output_tokens: NATIVE_COMPLETION_TOKEN_LIMIT },
             ...reasoningEffort ? {
               reasoning: {
                 effort: reasoningEffort,
@@ -4116,10 +3742,8 @@ Missing: ${missing.join("; ")}
             }
           }
           res = await this.postWithRetry(() => JSON.stringify({ ...body, stream: true, model: this.model }), call.signal, call);
-          if (!res) {
-            await this.ensureHostedModel();
+          if (!res)
             continue;
-          }
           if (!res.ok) {
             const responseText2 = await res.text();
             clearTimeout(timer);
@@ -4228,8 +3852,6 @@ Missing: ${missing.join("; ")}
         } else {
           throw new Error(`executeNative: provider ${this.provider} is not mapped to a wire`);
         }
-        if (!res)
-          await this.ensureHostedModel();
       } while (!res);
       const responseText = await res.text();
       clearTimeout(timer);
@@ -4507,9 +4129,14 @@ Missing: ${missing.join("; ")}
     };
     const decoder = new TextDecoder();
     let buffer = "";
+    let trailingCR = false;
+    let receivedBytes = 0;
+    let malformedEvents = 0;
+    const eventTypes = /* @__PURE__ */ new Set();
+    const identifier = (value) => typeof value === "string" && /^[A-Za-z0-9_.:-]{1,96}$/.test(value) ? value : null;
     let completedResponse = null;
-    let openRouterStreamFailed = false;
-    let openRouterUsage;
+    let streamFailure;
+    let responseUsage;
     const streamedOutputItems = [];
     let thinkingText = "";
     let lastThinkingEmit = 0;
@@ -4532,7 +4159,7 @@ Missing: ${missing.join("; ")}
       lastThinkingLength = thinkingText.length;
       callbacks.onThinking(thinkingText);
     };
-    while (true) {
+    responses: while (true) {
       let chunk;
       try {
         chunk = await readBounded();
@@ -4565,9 +4192,18 @@ Missing: ${missing.join("; ")}
         throw err;
       }
       const { done, value } = chunk;
-      if (done)
+      if (done) {
+        buffer += decoder.decode();
         break;
-      buffer += decoder.decode(value, { stream: true });
+      }
+      receivedBytes += value.byteLength;
+      let decoded = decoder.decode(value, { stream: true });
+      if (decoded) {
+        if (trailingCR && decoded.startsWith("\n"))
+          decoded = decoded.slice(1);
+        trailingCR = decoded.endsWith("\r");
+        buffer += decoded.replace(/\r\n?/g, "\n");
+      }
       let boundary = buffer.indexOf("\n\n");
       while (boundary >= 0) {
         const rawChunk = buffer.slice(0, boundary);
@@ -4580,24 +4216,42 @@ Missing: ${missing.join("; ")}
         let event;
         try {
           event = JSON.parse(payload);
+          if (!event || typeof event !== "object" || Array.isArray(event)) {
+            malformedEvents++;
+            continue;
+          }
         } catch {
+          malformedEvents++;
           continue;
         }
         const type = String(event.type ?? "");
-        if (this.provider === "openrouter" && (type === "response.done" || type === "response.failed" || type === "response.completed" || type === "response.incomplete")) {
-          const response = event.response;
-          const usage2 = response?.usage;
-          if (usage2 && typeof usage2.input_tokens === "number" && Number.isFinite(usage2.input_tokens) && usage2.input_tokens >= 0 && typeof usage2.output_tokens === "number" && Number.isFinite(usage2.output_tokens) && usage2.output_tokens >= 0) {
-            openRouterUsage = {
-              inputTokens: usage2.input_tokens,
-              outputTokens: usage2.output_tokens,
-              ...readResponsesCachedTokens(usage2)
+        if (eventTypes.size < 32)
+          eventTypes.add(identifier(type) ?? "unrecognized");
+        const terminal = type === "response.completed" || this.provider === "openrouter" && type === "response.done";
+        if (terminal || type === "response.failed" || type === "response.incomplete" || type === "error") {
+          const response = event.response && typeof event.response === "object" && !Array.isArray(event.response) ? event.response : void 0;
+          const usage = response?.usage;
+          if (usage && typeof usage.input_tokens === "number" && Number.isFinite(usage.input_tokens) && usage.input_tokens >= 0 && typeof usage.output_tokens === "number" && Number.isFinite(usage.output_tokens) && usage.output_tokens >= 0) {
+            responseUsage = {
+              inputTokens: usage.input_tokens,
+              outputTokens: usage.output_tokens,
+              ...readResponsesCachedTokens(usage)
             };
-            callbacks?.onUsage?.(openRouterUsage);
+            callbacks?.onUsage?.(responseUsage);
           }
-        }
-        if (this.provider === "openrouter" && (type === "error" || type === "response.failed" || type === "response.incomplete")) {
-          openRouterStreamFailed = true;
+          if (!terminal || !response || response.error != null || (type === "response.done" || response.status !== void 0) && response.status !== "completed") {
+            const error = response?.error ?? event.error;
+            const incomplete = response?.incomplete_details;
+            streamFailure = {
+              event: type,
+              status: identifier(response?.status),
+              code: identifier(error?.code ?? event.code),
+              errorType: identifier(error?.type),
+              reason: identifier(incomplete?.reason)
+            };
+            break responses;
+          }
+          completedResponse = response;
           continue;
         }
         if (type === "response.output_text.delta" || this.provider === "openrouter" && type === "response.content_part.delta") {
@@ -4631,33 +4285,33 @@ Missing: ${missing.join("; ")}
           }
           continue;
         }
-        if (type === "response.completed" || type === "response.incomplete" || this.provider === "openrouter" && type === "response.done") {
-          const response = event.response;
-          if (response) {
-            if (this.provider === "openrouter" && (openRouterStreamFailed || response.error != null || (type === "response.done" || response.status !== void 0) && response.status !== "completed")) {
-              openRouterStreamFailed = true;
-              continue;
-            }
-            completedResponse = response;
-            const usage2 = response.usage;
-            if (usage2 && this.provider !== "openrouter") {
-              callbacks?.onUsage?.({
-                inputTokens: Number(usage2.input_tokens ?? 0),
-                outputTokens: Number(usage2.output_tokens ?? 0)
-              });
-            }
-          }
-        }
       }
     }
     emitThinking(true);
-    if (!completedResponse || openRouterStreamFailed) {
+    if (!completedResponse || streamFailure) {
+      if (streamFailure) {
+        void reader.cancel().catch(() => {
+        });
+      }
+      appendNativeTrace({
+        kind: "native-response-stream-error",
+        provider: this.providerLabel,
+        wireApi: this.wireApi,
+        httpStatus: res.status,
+        eventStreamContentType: /^text\/event-stream(?:\s*;|$)/i.test(res.headers.get("content-type") ?? ""),
+        terminalFailure: streamFailure ?? null,
+        eventTypes: [...eventTypes],
+        receivedBytes,
+        malformedEvents,
+        trailingCharacters: buffer.length,
+        usage: responseUsage ?? null
+      });
       return {
         content: thinkingText ? [{ type: "text", text: thinkingText }] : [{ type: "text", text: "" }],
         stopReason: "error",
         durationMs: Date.now() - start,
-        ...openRouterUsage ? { usage: openRouterUsage } : {},
-        error: `${this.providerLabel} API error: ${openRouterStreamFailed ? "response stream failed" : "stream completed without final response"}`
+        ...responseUsage ? { usage: responseUsage } : {},
+        error: `${this.providerLabel} API error: ${streamFailure ? `Responses terminal failure ${JSON.stringify(streamFailure)}` : `stream completed without final response (HTTP ${res.status}; events=${[...eventTypes].join(",") || "none"}; malformed=${malformedEvents}; trailing=${buffer.length})`}`
       };
     }
     appendNativeTrace({
@@ -4710,21 +4364,10 @@ Missing: ${missing.join("; ")}
       thinkingText = reasoningSummaries.join("\n");
       emitThinking(true);
     }
-    const usageRecord = completedResponse.usage;
-    const usage = usageRecord ? {
-      inputTokens: Number(usageRecord.input_tokens ?? 0),
-      outputTokens: Number(usageRecord.output_tokens ?? 0),
-      // Responses `input_tokens` already INCLUDES the cached span (unlike
-      // Anthropic, which subtracts it), so no normalisation is needed —
-      // this is purely so cache behaviour becomes observable. Without it
-      // the Codex cache hit rate is unmeasurable: `prompt-cache.ts`
-      // instruments the Anthropic path only.
-      ...readResponsesCachedTokens(usageRecord)
-    } : void 0;
     return {
       content,
       stopReason: content.some((item) => item.type === "tool_use") ? "tool_use" : "end_turn",
-      usage,
+      usage: responseUsage,
       durationMs: Date.now() - start,
       // `outputItems` is the complete, correctly-ordered response array —
       // reasoning items with their `encrypted_content` still attached, each
@@ -4767,18 +4410,12 @@ export {
   isDiagnosticsClaimed,
   recentDiagnostics,
   _resetDiagnosticsForTests,
-  DEFAULT_CLOUD_HOST,
-  CloudAuthMissingError,
-  CloudAuthError,
-  loadCloudCredentials,
-  CloudError,
-  CloudUnauthorizedError,
-  CloudForbiddenError,
-  CloudNetworkError,
-  CloudClient,
   QuotaExhaustedError,
   OperatorAbortError,
   parseUsageLimitReached,
+  CODEX_CLIENT_VERSION,
+  CODEX_PROTOCOL_HEADERS,
   LOOP_SERVER_COMPACTION_TOKENS,
+  getChatGptCodexAccessToken,
   LlmApiRuntime
 };