audit-tools 0.34.42 → 0.35.1

package/docs/audit-pkg/operator-guide.md

@@ -181,119 +181,26 @@ Use `ui_mode: "visible"` when debugging provider stdout/stderr. Use
 `subprocess-template` or `vscode-task` only when you have a reliable launcher
 bridge.
 
-## Gate-0 provider confirmation
+## Relay-backed dispatch sources
 
-The first thing an interactive audit run does — before it reads the repo — is
-pause on a **provider confirmation** step (Gate-0). The tool auto-detects the
-provider pool it can dispatch to, prices each candidate, and proposes a cost
-ordering (cheapest capable first). You confirm, reorder, or amend it. The
-mechanism behind the ordering is documented in
-[`spec/dispatch-quota.md`](../../spec/dispatch-quota.md); this section is
-only how to respond to the step.
-
-### When it fires
-
-- On the conversation/CLI path (`audit-code next-step`) it fires on **every** run,
-  even when only one — or zero — providers were auto-detected, so you can reorder,
-  exclude, self-report your own model roster, or add a provider that discovery missed.
-- Headless (`audit-code advance-audit`, no interactive host) it auto-completes with
-  the tool's price-ascending suggestion and never pauses.
-
-### How to respond
-
-The step prints the suggested pool as a priced table and asks you to write your
-decision to:
-
-```text
-.audit-tools/audit/provider-confirmation.input.json
-```
-
-**Writing that file is what lets the run proceed** — its presence is the "operator
-has acted" signal. To accept the suggestion verbatim, write just the version:
-
-```json
-{ "schema_version": "provider-confirmation-input/v1" }
-```
-
-Every other field is optional:
+Audit-tools does not ask you to confirm provider/model order. Concrete candidates,
+ranking, cooldowns, and failover belong to the external broker. A machine-level
+`sources-declared.json` should expose only provider-neutral intents, for example:
 
 ```json
 {
-  "schema_version": "provider-confirmation-input/v1",
-  "cost_order": ["<provider-or-model-key>", "..."],
-  "exclude": ["transport:<name>", "transport:<name>/<model>", "service:<name>", "host:<endpoint-host>"],
-  "include": ["<self-spawn-blocked provider to opt back in>"],
-  "host_models": [{ "model_id": "<your model id>", "tier": "frontier|capable|fast" }]
+  "sources": [
+    { "id": "relay-fast", "transport": "openai-compatible", "endpoint": "http://127.0.0.1:8791/v1", "model": "pool/fast", "no_auth": true, "worker_kind": "single_shot", "capability_rank": 2 },
+    { "id": "relay-coding", "transport": "openai-compatible", "endpoint": "http://127.0.0.1:8791/v1", "model": "pool/coding", "no_auth": true, "worker_kind": "single_shot", "capability_rank": 1 },
+    { "id": "relay-reasoning", "transport": "openai-compatible", "endpoint": "http://127.0.0.1:8791/v1", "model": "pool/reasoning", "no_auth": true, "worker_kind": "single_shot", "capability_rank": 0 }
+  ]
 }
 ```
 
-- `cost_order` — the confirmed ordering, cheapest first, as a list of keys. A key
-  is either a **provider name** (as shown in the table) or a **host `model_id`** you
-  report in `host_models`. Keys you omit keep their suggested relative order,
-  appended after the ones you name; unrecognized keys are ignored. Omit it to accept
-  the suggested order.
-- `exclude` — drops backends from the dispatchable pool. Each entry names its AXIS
-  explicitly, so the same string means the same thing on every machine. The grammar
-  is defined in [`spec/backend-identity-axes.md`](../../spec/backend-identity-axes.md)
-  → *The exclusion grammar: axis-explicit*; what follows is how to use it, and that
-  spec wins on any disagreement:
-  - `transport:openai-compatible/gpt-oss-120b` — that **model** of that transport,
-    leaving its other models routable. This is the granularity to reach for: you are
-    confirming model choices, so ruling out one model of a multi-model backend
-    should not rule out the backend.
-  - `transport:codex` — the whole **transport**, every model of it.
-  - `service:nim` — every transport reaching that **service**, including ones
-    discovered later.
-  - `host:integrate.api.nvidia.com` (or `host:localhost:8000` to pin a port) — every
-    source at that **endpoint host**.
-
-  The older prefix-less forms (`codex`, `openai-compatible:gpt-oss-120b`,
-  `integrate.api.nvidia.com`) are still accepted and migrated to the axis form on
-  read, so saved rules keep working. A rule naming an axis that does not exist —
-  `model:gpt-oss-120b` — is REFUSED when you submit it, rather than being silently
-  reinterpreted as a host rule that matches nothing.
-
-  A rule is a durable decision, not a reachability claim: it keeps applying on a
-  later run and on a different machine, whatever is reachable there. A rule that
-  matches nothing is legitimate and stays inert — the tool authors one itself every
-  run — but the confirmation summary lists any of YOUR rules that matched nothing,
-  as an advisory, so a model-name typo is visible without blocking anything.
-- `include` — opts a self-spawn-blocked provider back in. A CLI provider detected
-  while you are already inside a session of that same agent (`claude-code` under
-  `CLAUDECODE`, `codex` under `CODEX`) is excluded by default so the run can't
-  self-spawn a fresh agent; naming it here overrides that. Advanced — normally leave
-  it excluded.
-- `host_models` — reports your own (the host agent's) model roster so those tiers
-  are priced from the models.dev dataset and orderable here, not just at dispatch.
-
-You supply only ordering intent plus your model roster. The tool owns the prices
-and the capability flags — you never hand-author those. If a
-provider you use isn't listed, it wasn't auto-detected: it rides the per-auditor
-`--auditor <json>` descriptor (an OpenAI-compatible endpoint or CLI backend as a
-`sources[]` entry, resolved from your environment — NOT `session-config.json`, which
-no longer accepts dispatch config; see `spec/unified-dispatch-worker-model.md`), then
-re-run the step.
-
-Once the input is written, re-run the continue command the step printed
-(`audit-code next-step`). The tool consumes the input and promotes it into both
-canonical artifacts: the per-tool `provider_confirmation.json` seam and the shared
-`.audit-tools/provider-confirmation.json`.
-
-### The confirmed pool persists — you aren't re-prompted
-
-The shared `.audit-tools/provider-confirmation.json` is written once and reused for
-the rest of the audit→remediate session. A later step, and a subsequent
-`remediate-code` run against the same repo, read and honor that pool without
-prompting again — `remediate-code` has no confirmation step of its own; it consumes
-the audit-side pool. You are re-prompted only when a backend you never confirmed
-becomes reachable — and then the prompt leads with just that delta, not the whole
-table again. A backend that has *disappeared* never re-prompts: it simply drops out
-of reach, which needs no decision from you. On an unattended run (`autonomous_mode`)
-there is nobody to ask, so a newly-reachable backend is **excluded rather than used**
-and a `newly_reachable_backend` friction event records why — confirm or exclude it at
-your next attended gate. If the file is absent
-(for example a standalone remediate run with no prior audit), the run resolves its
-provider independently — absence is never an error.
+The relay owns the concrete model lists behind those pool names. Audit-tools still
+enforces packet size/context fit, capability floors, quota/headroom, concurrency,
+result validation, and the local self-spawn guard. No provider-order input or
+provider-confirmation artifact is created.
 
 ## Model selection
 

package/docs/audit-pkg/product.md

@@ -38,17 +38,18 @@ The intended workflow is:
 
 1. The user invokes `/audit-code`.
 2. The prompt runs `audit-code ensure --quiet`.
-3. The conversation confirms (or accepts by default) the discovered provider
-   pool and cost ordering (`provider_confirmation`) — the pipeline's first
-   gated step.
-4. Deterministic backend steps build or refresh artifacts.
-5. The active conversation dispatches bounded review packets when semantic
+3. Deterministic backend steps build or refresh artifacts.
+4. The active conversation dispatches bounded review packets when semantic
    judgment is required.
-6. Packet workers submit validated `AuditResult` objects through backend-owned
+5. Packet workers submit validated `AuditResult` objects through backend-owned
    commands.
-7. The backend ingests results, performs selective deepening and runtime
+6. The backend ingests results, performs selective deepening and runtime
    validation when needed, and writes the final `audit-report.md`.
 
+Concrete provider/model ordering and failover belong to the external dispatch
+broker. The audit workflow consumes provider-neutral pool intents and never asks
+the user to confirm a provider roster.
+
 Semantic review belongs to the active host conversation by default. The backend
 provider adapters — the authoritative current roster is `PROVIDER_NAMES`
 (`src/shared/types/sessionConfig.ts`) — are compatibility bridges for

package/docs/audit-pkg/release.md

@@ -9,30 +9,41 @@ npm ci
 npm run verify:release
 ```
 
-`verify:release` covers:
+`verify:release` runs `verify:checks` first, then the full vitest suite, then both
+linked-install smokes. The `verify:checks` half is generated from `package.json`
+below — in its real order, so this list cannot drift from the gate:
+
+<!-- BEGIN gate-enumeration — generated from package.json by scripts/check-gate-enumeration.mjs -->
 
-- TypeScript typecheck (`build`)
-- test-tree typecheck (`check:tests`)
 - raw control-byte gate (`check:control-bytes`)
 - version-gate scan (`check:version-gates`)
+- guard wiring/reach reconciliation (`check:guard-reach`)
 - loop-core pattern-list drift check (`check:loop-core-patterns`)
 - constitutional-doc-path parity (`check:constitutional-doc-paths`)
-- backlog size-budget gate (`check:backlog-budget`)
-- backlog status-token gate (`check:backlog-status`)
-- backlog seek-index parity (`check:backlog-index`)
-- memory-citation check (`check:memory-citations`)
-- README philosophy-brief parity (`check:philosophy-brief`)
-- nightly scheduler-prompt parity (`check:nightly-routine-prompt`)
-- HANDOFF roadmap parity (`check:handoff-roadmap`)
 - dead-code export gate (`check:deadcode`)
 - doc-manifest reconciliation gate (`check:doc-manifest`)
 - relative-link resolution gate (`check:doc-links`)
 - backticked repo-path citation gate (`check:doc-code-citations`)
+- gate-enumeration parity (this list) (`check:gate-enumeration`)
+- README philosophy-brief parity (`check:philosophy-brief`)
+- nightly scheduler-prompt parity (`check:nightly-routine-prompt`)
+- HANDOFF roadmap parity (`check:handoff-roadmap`)
+- backlog seek-index parity (`check:backlog-index`)
+- memory-citation check (`check:memory-citations`)
+- backlog size-budget gate (`check:backlog-budget`)
+- backlog status-token gate (`check:backlog-status`)
+- test-tree typecheck (`check:tests`)
+- TypeScript typecheck (`build`)
+- host-install verification (audit) (`verify:hosts`)
+- host-install verification (remediate) (`verify:remediate-hosts`)
 - single-tarball pack smoke (`pack:smoke`)
-- full automated test suite (`vitest run`)
-- host-install verification for both bins (`verify:hosts`, `verify:remediate-hosts`)
-- packaged-install smoke coverage for both bins (`smoke:packaged-audit-code`, `smoke:packaged-remediate-code`)
-- linked-install smoke coverage for both bins (`smoke:linked-audit-code`, `smoke:linked-remediate-code`)
+- packaged-install smoke (audit-code) (`smoke:packaged-audit-code`)
+- packaged-install smoke (remediate-code) (`smoke:packaged-remediate-code`)
+
+<!-- END gate-enumeration -->
+
+…followed by the full automated test suite (`vitest`), then
+`smoke:linked-audit-code` and `smoke:linked-remediate-code`.
 
 For live child-process output while debugging smoke tests:
 
@@ -61,21 +72,19 @@ no-drift guard does not see that, so a human runs these rows at release. A faile
 row becomes a backlog item.
 
 Codex and `agy` are headless CLIs, so they are correctly absent from the GUI-host
-table below and their live dispatch is automated instead of listed here — see the
-`RUN_PROVIDER_MATRIX_E2E=1`-gated provider-matrix e2e in
-`tests/audit/provider-matrix-dispatch-e2e.test.ts` (supersedes the former
-per-provider `RUN_CODEX_E2E=1` gate). Run it at release with a live backend
-present:
-
-```
-RUN_PROVIDER_MATRIX_E2E=1 npx vitest run tests/audit/provider-matrix-dispatch-e2e.test.ts
-```
-
-Coverage gap: that e2e currently exercises only `codex` / `opencode` /
-`openai-compatible`. `agy` and `claude-worker` have **no live-dispatch e2e coverage**
-yet — until a row is added for each, their real dispatch is unverified by any
-automated gate (`claude-worker`'s existing test, `tests/shared/claude-worker-provider.test.ts`,
-is a local-mock-HTTP-server unit test of transport/argv/env, not a live-dispatch e2e).
+table below. The `RUN_PROVIDER_MATRIX_E2E=1`-gated provider-matrix live e2e (and
+its `runInProcessAuditDispatch` driver) was **retired with the dispatch
+inversion**: the host and llm-relay own concrete provider/model routing now, so
+audit-tools no longer selects or instantiates those backends for host review and
+the per-backend live round-trip it exercised is no longer this package's seam.
+
+Coverage gap, stated so it is not mistaken for coverage: **no headless backend has
+live-dispatch e2e coverage** — worker-side dispatch through a configured source
+(openai-compatible / codex / agy / `claude-worker`) is verified by unit tests over
+local stubs only (e.g. `tests/shared/claude-worker-provider.test.ts` is a
+local-mock-HTTP-server unit test of transport/argv/env, and
+`tests/remediate/hybrid-decision-point-pools.test.ts` drives a local stub
+endpoint), never by a gate against a live backend.
 
 ### How to run a row
 

package/opencode.json

@@ -8,9 +8,9 @@
     "grep": "allow",
     "edit": {
       "*": "ask",
-      ".audit-code/**": "allow",
-      ".audit-tools/**": "allow",
       ".remediate-code/**": "allow",
+      ".audit-tools/**": "allow",
+      ".audit-code/**": "allow",
       ".audit-artifacts/**": "allow",
       "audit-report.md": "allow"
     },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "audit-tools",
-  "version": "0.34.42",
+  "version": "0.35.1",
   "private": false,
   "license": "ISC",
   "description": "Portable hybrid code auditing + remediation orchestrators for arbitrary repositories.",
@@ -42,12 +42,13 @@
     "build": "tsc -p tsconfig.json && node scripts/shared/copy-data-assets.mjs",
     "check": "tsc -p tsconfig.json --noEmit",
     "check:tests": "tsc -p tsconfig.test.json",
-    "nightly:review": "node scripts/nightly/serve.mjs",
+    "nightly:inbox": "node scripts/nightly/render-inbox.mjs",
+    "nightly:ingest": "node scripts/nightly/ingest-answers.mjs",
     "test:doc-contract": "node scripts/shared/run-vitest-gate.mjs tests/audit/release-contract.test.ts tests/audit/file-lock-doc-sync.test.ts tests/audit/host-asset-renderer-drift.test.ts",
     "test": "npm run build && node scripts/shared/run-vitest-gate.mjs",
     "test:single": "npm run build && node scripts/shared/run-vitest-gate.mjs",
     "verify:guards": "node scripts/shared/run-vitest-gate.mjs --retry=2 --exclude \"**/audit-code-completion.test.ts\" --exclude \"**/audit-code-wrapper.test.ts\" --exclude \"**/audit/next-step.test.ts\" --exclude \"**/seam-host-only-next-step.test.ts\" --exclude \"**/*-e2e.test.*\" --exclude \"**/next-step-implement-dispatch.test.ts\" --exclude \"**/next-step-pipeline-dispatch.test.ts\"",
-    "verify:checks": "node scripts/shared/profile-run.mjs verify-checks check:control-bytes check:version-gates check:guard-reach check:loop-core-patterns check:constitutional-doc-paths check:deadcode check:doc-manifest check:doc-links check:doc-code-citations check:philosophy-brief check:nightly-routine-prompt check:handoff-roadmap check:backlog-index check:memory-citations check:backlog-budget check:backlog-status check:tests build verify:hosts verify:remediate-hosts pack:smoke smoke:packaged-audit-code smoke:packaged-remediate-code",
+    "verify:checks": "node scripts/shared/profile-run.mjs verify-checks check:control-bytes check:version-gates check:guard-reach check:loop-core-patterns check:constitutional-doc-paths check:deadcode check:doc-manifest check:doc-links check:doc-code-citations check:gate-enumeration check:philosophy-brief check:nightly-routine-prompt check:handoff-roadmap check:backlog-index check:memory-citations check:backlog-budget check:backlog-status check:tests build verify:hosts verify:remediate-hosts pack:smoke smoke:packaged-audit-code smoke:packaged-remediate-code",
     "verify:release": "npm run verify:checks && node scripts/shared/run-vitest-gate.mjs && npm run smoke:linked-audit-code && npm run smoke:linked-remediate-code",
     "check:control-bytes": "node scripts/check-control-bytes.mjs",
     "check:guard-reach": "node scripts/check-guard-reach.mjs",
@@ -84,7 +85,8 @@
     "check:handoff-roadmap": "node scripts/shared/generate-handoff-roadmap.mjs --check",
     "check:backlog-status": "node scripts/check-backlog-status-tokens.mjs",
     "check:backlog-index": "node scripts/shared/generate-backlog-index.mjs --check",
-    "test:rebaseline-flakes": "node scripts/shared/rebaseline-flakes.mjs"
+    "test:rebaseline-flakes": "node scripts/shared/rebaseline-flakes.mjs",
+    "check:gate-enumeration": "node scripts/check-gate-enumeration.mjs"
   },
   "engines": {
     "node": ">=20"

package/skills/audit-code/SKILL.md

@@ -63,7 +63,8 @@ Bounded steps are a backend implementation detail, not the intended user experie
 After `audit-code ensure --quiet` bootstraps the repository, ask the backend for
 exactly one next step. This is also the **capability handshake**: report what you
 can dispatch to *right now* on every `next-step` call so the backend sizes review
-packets to your real model instead of a conservative 32k floor.
+packets against reported or configured limits. When no authoritative limits are
+available, capacity remains unknown and packet sizing pauses.
 
 The whole handshake is a single `--auditor <json>` flag: one object with a `self`
 field describing the model you drive.

package/skills/audit-code/audit-code.prompt.md

@@ -26,9 +26,10 @@ node audit-code.mjs ensure --quiet
 
 Then ask the backend for exactly one next step. This is also the **capability
 handshake**: report what you can dispatch to *right now* on every `next-step`
-call as a single `--auditor <json>` flag, so the backend sizes review packets to
-your real model instead of a conservative 32k floor. The JSON is one object with
-a `self` field describing the model you drive; report inside `self`:
+call as a single `--auditor <json>` flag, so the backend sizes review packets
+against reported or configured limits. Without them, capacity remains unknown
+and packet sizing pauses. The JSON is one object with a `self` field describing
+the model you drive; report inside `self`:
 
 - `roster` — an ordered JSON array (lowest rank first) of the models you can
   dispatch subagents to *right now*, one entry per relative rank:

package/skills/remediate-code/remediate-code.prompt.md

@@ -53,7 +53,8 @@ remediate-code next-step
 
 Every `next-step` call is also the **capability handshake**: report what you can
 dispatch to *right now* so the backend sizes remediation waves and per-worker
-context to your real model instead of a conservative 32k floor. Report:
+context against reported or configured limits. Without them, capacity remains
+unknown and wave planning pauses. Report:
 
 - `--host-can-dispatch-subagents` — whether you can run callable subagents at all
   (via the `Agent`/`task` tool). Without it the backend runs remediation serially.

package/dist/audit/cli/providerConfirmationStep.d.ts

@@ -1,106 +0,0 @@
-import type { ConfirmedPoolEntry, SourcePoolDisplayEntry, NewlyReachableBackend } from "audit-tools/shared";
-/**
- * Render the host-facing prompt for the interactive `provider_confirmation` step
- * (Gate-0, cost-first routing — spec/dispatch-quota.md). Shows the priced,
- * suggested provider pool (ascending cost) and asks the host to confirm it, or to
- * submit an operator reorder / exclusion / host-model roster by writing
- * `provider-confirmation.input.json`. Writing that file (even to accept the
- * suggestion verbatim) is what lets the run proceed — its presence is the
- * "operator has acted" signal the gate consumes.
- *
- * The tool owns discovery + the cost annotation: the host supplies only ordering
- * intent + its model roster, and the tool promotes that into both canonical
- * artifacts. So this prompt never asks the host to hand-author prices or
- * capability flags.
- */
-export declare function renderProviderConfirmationPrompt(opts: {
-    /** The tool's suggested pool, already annotated with model_id/price/cost_order. */
-    providerPool: ConfirmedPoolEntry[];
-    /**
-     * Every configured `sessionConfig.sources[]` pool (backlog follow-up a) —
-     * display-only, derived by `deriveSourcePoolDisplay`. These are NOT part of
-     * `providerPool` / its `cost_order` sequence, but `collectDispatchableSources`
-     * (apiPool.ts) folds every one of them into dispatch, so they must be visible
-     * here or the operator is confirming an ordering that omits pools that WILL
-     * route. Omit/empty when no `sources[]` are configured.
-     */
-    sourcePools?: SourcePoolDisplayEntry[];
-    /** Absolute path the host writes its operator input to. */
-    inputPath: string;
-    /** The exact command to re-invoke once the input is written. */
-    continueCommand: string;
-    /**
-     * The G3 reconciliation delta: backends reachable now that the operator's prior
-     * confirmation never mentioned. Non-empty ⇒ this is a RE-confirmation and the
-     * prompt leads with the delta, so the operator answers the new question instead
-     * of re-reading a whole table they already approved. Empty/absent ⇒ the ordinary
-     * first-time gate.
-     */
-    newlyReachable?: readonly NewlyReachableBackend[];
-    /**
-     * The capability-evidence delta: dispatchable models with no resolvable capability
-     * rank. Non-empty ⇒ the prompt additionally asks for a most-capable-first ordering
-     * over exactly these models, which persists as `capability_rank` and stops them
-     * fail-opening at the admission capability floor.
-     */
-    unevidencedCapability?: readonly string[];
-    /**
-     * A BOUNDED, spread sample of the already-confirmed capability ordering
-     * (`selectCapabilityAnchors`), most-capable-first — the fixed reference points the
-     * operator ranks the unevidenced models against.
-     *
-     * Without them the capability ask is unanswerable without livelocking: the prompt is
-     * delta-scoped, so an answer over the new models alone used to REPLACE the confirmed
-     * ordering and the models it dropped came straight back as the next delta. Rendering
-     * the whole ordering instead is not an option — the roster may be hundreds of models
-     * and this prompt must stay O(new + constant). Empty on a first-ever ranking.
-     */
-    capabilityAnchors?: readonly string[];
-    /**
-     * True ⇒ a prior confirmation exists on disk, i.e. this is a RE-confirmation.
-     *
-     * Drives the do-not-re-litigate guardrail INDEPENDENTLY of the reach delta. It used to
-     * ride on `newlyReachable`, so a capability-only re-prompt rendered the full table with
-     * no warning at all — while also telling the operator that an omitted field keeps its
-     * confirmed value. An operator transcribing the displayed ordering into `cost_order`
-     * then silently reverted their own confirmed order.
-     */
-    hasPriorConfirmation?: boolean;
-    /**
-     * R3-3: true on an autonomous (no-operator) run. Only changes rendering when
-     * `unevidencedCapability` is non-empty — addresses that section to the HOST LLM
-     * (an autonomous run has no operator to ask, but the CLI's `next-step` prompt is
-     * always executed by a host agent, so THAT is who ranks) instead of an operator,
-     * and OMITS the reach-reconciliation section entirely: reach is never an LLM's
-     * call to make (`intakeExecutors.ts`'s executor fails it closed regardless of
-     * what an LLM-authored submission says), so the prompt must not ask one to answer
-     * a question it must not answer. Absent/false renders the ordinary operator
-     * variant, unchanged.
-     */
-    autonomous?: boolean;
-    /**
-     * R3-3: anchor ids whose current position is LLM-authored rather than
-     * operator-authored (`ConfirmedDispatchPolicy.capability_order_llm_ranked`).
-     * Drives the ATTENDED variant's anchor marker only — an operator sees
-     * "(LLM-ranked — restate it in your order to reposition it)" instead of
-     * "(already ranked)" for these, since (unlike a genuinely operator-ranked
-     * anchor) they may freely reposition one without restating the whole roster.
-     * Has no effect on the autonomous variant (the LLM must not reorder ANY anchor,
-     * regardless of who ranked it) or when the id is not among `capabilityAnchors`.
-     */
-    capabilityOrderLlmRanked?: readonly string[];
-    /**
-     * Declared `sources[]` entries that did NOT resolve this invocation, with the
-     * operator-facing reason each carries (`resolveAmbientSources`' `dropped[]`).
-     *
-     * Rendered so the drop reaches the operator in the GATE-0 PROMPT, which is the
-     * artifact they actually read. Its only other channel is a stderr line, which
-     * conversation-first use never surfaces — leaving a declared lane silently
-     * missing from the table with no stated cause.
-     */
-    droppedSources?: readonly {
-        id: string;
-        reason: string;
-    }[];
-}): string;
-//# sourceMappingURL=providerConfirmationStep.d.ts.map

package/dist/audit/cli/providerConfirmationStep.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"providerConfirmationStep.d.ts","sourceRoot":"","sources":["../../../src/audit/cli/providerConfirmationStep.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,kBAAkB,EAClB,sBAAsB,EACtB,qBAAqB,EACtB,MAAM,oBAAoB,CAAC;AAG5B;;;;;;;;;;;;;GAaG;AACH,wBAAgB,gCAAgC,CAAC,IAAI,EAAE;IACrD,mFAAmF;IACnF,YAAY,EAAE,kBAAkB,EAAE,CAAC;IACnC;;;;;;;OAOG;IACH,WAAW,CAAC,EAAE,sBAAsB,EAAE,CAAC;IACvC,2DAA2D;IAC3D,SAAS,EAAE,MAAM,CAAC;IAClB,gEAAgE;IAChE,eAAe,EAAE,MAAM,CAAC;IACxB;;;;;;OAMG;IACH,cAAc,CAAC,EAAE,SAAS,qBAAqB,EAAE,CAAC;IAClD;;;;;OAKG;IACH,qBAAqB,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IAC1C;;;;;;;;;;OAUG;IACH,iBAAiB,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IACtC;;;;;;;;OAQG;IACH,oBAAoB,CAAC,EAAE,OAAO,CAAC;IAC/B;;;;;;;;;;OAUG;IACH,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB;;;;;;;;;OASG;IACH,wBAAwB,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IAC7C;;;;;;;;OAQG;IACH,cAAc,CAAC,EAAE,SAAS;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,EAAE,CAAC;CAC5D,GAAG,MAAM,CA8YT"}