pentesting 0.100.4 → 0.100.5

package/ARCHITECTURE.md

@@ -27,6 +27,9 @@ stays intentionally short and usage-focused.
   access (policy-based today; OS sandboxing is roadmap).
 - **Intelligent Queuing** — Queued inputs are deduplicated and prioritized live;
   obsolete tasks are preempted, not FIFO-replayed.
+- **Prompt Ownership** — Prompt mode owns Enter submission. Buffered replay may
+  preserve typed-ahead input, but live terminal control responses, redraws, and
+  resize noise are not allowed to turn a normal Enter into a pasted newline.
 - **Lab Session Multiplexing** — `shell-listener` runs multiple authorized TCP
   sessions with per-session routing, identity tags, verified PTY upgrade state,
   transcripts, command ledgers, replay, and orphaned evidence recovery.
@@ -66,6 +69,7 @@ Plain view:
 ```text
 user input
   -> CLI / interactive prompt loop
+  -> terminal-session prompt ownership + buffered replay boundary
   -> BuilderAPI facade
   -> BuilderApp classification and strategy
   -> active agent or delegated subagent
@@ -85,6 +89,27 @@ There is no fixed persona. The runtime derives a profile per request, optionally
 overlays a named autonomy profile, then reuses tool scope, phase, and memory
 weighting.
 
+CTF frontier exploration keeps the global `[exploration]` default off, then
+enables the effective CTF profile when `engagement.kind = "ctf"` or a
+`flag_format` is configured. That profile selects `ctf-competition` autonomy,
+turns on exploration, requires verified `flag_check` evidence before completion,
+and auto-continues the TUI loop after `/goal`. See `docs/config/ctf.toml`,
+`docs/plans/2026/07/08/PLAN_CtfFrontierExploration_2026-07-08.md`, and the
+decision record beside it.
+
+The shipped design is advisor-guided, not scheduler-owned: `decide_next` is a
+pure traversal recommendation over `EvidenceKind::Lead` events, while dispatch
+uses the existing delegated-task fan-out. There is no detached task registry or
+second runtime scheduler in the default implementation. Phase 10 is pinned by
+`orch_spec::exploration_e2e`, which exercises the flag-off path and the
+breadth/depth/pivot/reseed/merge flow without launching real subagents.
+
+Reliability guards sit below the profile layer. Provider protocol capabilities
+default native tool support for OpenAI/Anthropic-compatible responses on
+registry misses, unparsed tool-shaped markup becomes retryable instead of a
+false completion, raw control markup is scrubbed from the stream, and the
+markdown renderer bounds very large no-newline buffers.
+
 ```mermaid
 flowchart LR
     R[Request] --> GEN["Derive<br/>dynamic profile"]
@@ -145,8 +170,9 @@ definition under `crates/builder_repo/src/agents/`.
 ## Named Autonomy Profiles
 
 Optionally pin the top-level profile (`autonomy_profile = "ctf-competition"`);
-leave unset for classifier-driven defaults. Delegated subagent contracts stay
-intact. Defined in `crates/builder_domain/src/autonomy_profile.rs`.
+leave unset for classifier-driven defaults. CTF engagement config auto-selects
+the CTF profile when no explicit profile is set. Delegated subagent contracts
+stay intact. Defined in `crates/builder_domain/src/autonomy_profile.rs`.
 
 | Profile | Purpose |
 | --- | --- |
@@ -157,9 +183,10 @@ intact. Defined in `crates/builder_domain/src/autonomy_profile.rs`.
 
 ## Engagement Metadata
 
-Runs can attach typed engagement context — scope, phase, tags, and standard refs
-(PTES, MITRE ATT&CK, OWASP, CWE/CAPEC, NIST CSF, CIS Controls) — to workflow
-metadata. `/workflow report` exports it as a Markdown handoff.
+Runs can attach typed engagement context — scope, phase, tags, flag format,
+flag-evidence requirement, and standard refs (PTES, MITRE ATT&CK, OWASP,
+CWE/CAPEC, NIST CSF, CIS Controls) — to workflow metadata. `/workflow report`
+exports it as a Markdown handoff.
 
 ## Memory & Knowledge
 
@@ -316,14 +343,14 @@ npm install -g pentesting
 
 ## Supported Runtime Targets
 
-The npm launcher currently resolves managed native release assets for the
-targets below. Other operating systems can run the Docker image or use
+The npm launcher currently resolves a managed native release asset for Linux
+x64. Other operating systems, including Apple Silicon Macs, Linux arm64,
+Windows, and Android, can run the multi-architecture Docker image or use
 `PENTESTING_BIN` to point at a locally built compatible binary.
 
 | OS | CPU | Release asset |
 | --- | --- | --- |
 | Linux | x64 | `pentesting-x86_64-unknown-linux-musl` |
-| Android | arm64 | `pentesting-aarch64-linux-android` |
 
 `/update` (and the npm postinstall) resolves the matching asset for the current
 supported OS/CPU target from `agnusdei1207/pentesting-public`.

package/README.md

@@ -25,10 +25,11 @@ Agent-loop essentials plus a security-focused runtime layer.
 | Tool guardrails | Read-before-edit enforcement, repeat-call guard, policy-gated file/shell/network |
 | Permissions | Allow / Deny / Confirm, with once/always decisions persisted |
 | Context compaction | Automatic micro-pruning + model-triggered summarization, pre/post hooks |
-| Interrupt & steering | ESC to interrupt mid-turn, queue follow-ups while it works |
+| Interrupt & steering | ESC to interrupt mid-turn, queue follow-ups while it works, and keep Enter submissions deterministic across PTY redraw/control-response noise |
 | Reverse-shell control plane | Accept and track multiple authorized callbacks, tag/probe sessions, send/observe/run commands, verify PTY upgrades, and keep transcript/ledger/evidence records |
 | PTY session control | Treat upgraded shells as managed PTY sessions with resize gating, transcript capture, FD accounting, and deterministic close/revoke/gc reclamation |
-| Subagents | Parallel delegated tasks, budget-capped, role-specific personas |
+| Frontier-guided exploration | Persistent, deduplicated lead frontier: the runtime enforces pivots and caps, and recommends breadth-first enumeration vs depth-first exploitation while the model drives |
+| Subagents | Broad-clone delegates with category focus overlays; independent deep-specialist leads (crypto/rev/pwn) run in the same parallel batch so the main line keeps moving |
 | MCP & skills | Dynamic MCP tool discovery, layered markdown skills |
 | Memory | Ebbinghaus-style decaying notes + hybrid lexical/semantic/graph retrieval |
 
@@ -68,6 +69,10 @@ docker run -it --rm \
   agnusdei1207/pentesting:latest
 ```
 
+Provider environment variables are optional for startup. If they are omitted,
+the interactive TUI still launches and lets you configure/login inside the
+persistent `pentesting-config` volume.
+
 Or via Docker Compose:
 
 ```bash
@@ -134,6 +139,46 @@ export PENTESTING_SKIP_DOWNLOAD=1                        # skip postinstall bina
 
 The TUI pre-fills output max tokens from known model names/provider metadata. `OPENAI_MAX_TOKENS` and the TUI field are explicit overrides for custom/self-hosted models.
 
+### Frontier-guided CTF exploration
+
+The global frontier engine default stays off, but CTF-shaped engagement config
+auto-applies the CTF runtime profile: `ctf-competition` autonomy, runtime
+exploration, autonomous continuation, and a completion gate that requires
+verified `flag_check` evidence. A ready-to-copy profile lives at
+`docs/config/ctf.toml`:
+
+```toml
+[engagement]
+kind = "ctf"
+flag_format = "flag\\{[^}]+\\}"
+require_flag_evidence = true
+
+[exploration]
+enabled = true
+max_frontier_nodes = 256
+max_breadth = 4
+```
+
+With the profile active, the runtime stores `explore` leads as
+`EvidenceKind::Lead` events, carries prior-run frontier/findings/verified flags
+forward, deduplicates leads, auto-pivots dead probes, enforces graph/cap limits,
+and injects a breadth/depth recommendation into mission control. The model still
+drives the work through `explore` and `task`; heavy independent leads such as
+crypto/rev/pwn ride the same parallel delegation batch rather than a detached
+background scheduler. Operators can inspect the current graph with `/frontier`,
+and the status row shows compact frontier progress during dispatch.
+
+Malformed tool-call output is fail-loud: unparsed tool-shaped markup is retried
+instead of accepted as a final answer, the TUI suppresses raw control markup, and
+large no-newline streams are bounded so rendering cannot hang behind usage-only
+progress ticks.
+
+The shipped contract is covered by
+`cargo test -p builder_app orch_spec::exploration_e2e`: it verifies the
+flag-off regression path plus breadth enumeration, depth descent, sibling
+breadth, dead-probe pivot, global reseed, seed digest, broad-clone overlay, and
+child result promotion.
+
 With Docker, mount both your project and the persistent runtime config volume, then pass the same variables through with `-e`:
 
 ```bash
@@ -148,6 +193,10 @@ docker run -it --rm \
   agnusdei1207/pentesting:latest
 ```
 
+Those variables only pre-fill provider settings. Leaving them unset must not
+block the container from launching the TUI; configure the provider interactively
+inside the container instead.
+
 > **Note:** The command you run is always **`pentesting`**. The internal engine is **`builder`** — `pentesting` downloads and runs it under the hood; the engine name never surfaces in normal use.
 
 ---

package/lib/runtime.mjs

@@ -14,15 +14,9 @@ const MANAGED_BINARY_DIR = path.join(PACKAGE_ROOT, "vendor");
 const MANAGED_BINARY_MANIFEST = path.join(MANAGED_BINARY_DIR, "pentesting-manifest.json");
 
 const RELEASE_TARGETS = {
-  // Native binaries published as release assets. Linux uses musl (static, covers
-  // x64/arm64); android is the cross-built target. macOS/Windows are not
-  // cross-buildable from the Linux release host, so they are served by the
-  // Docker image (see the Unsupported-platform error below).
-  "android:arm64": {
-    assetName: "pentesting-aarch64-linux-android",
-    binaryFileName: "builder",
-    label: "android-arm64",
-  },
+  // Native binaries published as release assets. Linux x64 uses musl (static).
+  // macOS, Windows, Linux arm64, and Android are served by the multi-arch Docker
+  // image (see the unsupported-platform error below).
   "linux:x64": {
     assetName: "pentesting-x86_64-unknown-linux-musl",
     binaryFileName: "builder",
@@ -60,7 +54,7 @@ export function resolveReleaseAsset(platform = process.platform, arch = process.
     throw new Error(
         `No native pentesting binary for '${platform}/${arch}'. ` +
         `Native targets: ${supported}. ` +
-        `On macOS/Windows run it via Docker instead: ` +
+        `On macOS/Windows/Linux arm64/Android run it via Docker instead: ` +
         `docker run -it --rm -v "$(pwd):/workspace" -v pentesting-config:/root/.pentesting -w /workspace -e OPENAI_API_KEY -e OPENAI_BASE_URL -e OPENAI_MODEL -e OPENAI_MAX_TOKENS agnusdei1207/pentesting`,
     );
   }

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "pentesting",
-  "version": "0.100.4",
-  "builderReleaseTag": "v0.100.1",
+  "version": "0.100.5",
+  "builderReleaseTag": "v0.100.5",
   "description": "pentesting — coding agent runtime with audited reverse-shell capture, verified PTY session upgrades, pivot control, evidence transcripts, and safe session reclamation.",
   "license": "MIT",
   "author": "agnusdei1207",
@@ -72,7 +72,7 @@
     "public:sync": "bash scripts/sync-public-repo.sh",
     "public:mirror-release": "bash scripts/mirror-public-release.sh",
     "release:backfill": "bash scripts/backfill-release-local.sh",
-    "check:docker": "sh -c ': \"${OPENAI_API_KEY:?Set OPENAI_API_KEY}\"; : \"${OPENAI_BASE_URL:?Set OPENAI_BASE_URL}\"; : \"${OPENAI_MODEL:?Set OPENAI_MODEL}\"; : \"${OPENAI_MAX_TOKENS:?Set OPENAI_MAX_TOKENS}\"; npm run docker:build && docker run -it --rm -v builder-workspace:/workspace -v pentesting-config:/root/.pentesting -e OPENAI_API_KEY -e OPENAI_BASE_URL -e OPENAI_MODEL -e OPENAI_MAX_TOKENS agnusdei1207/pentesting:latest'",
+    "check:docker": "npm run docker:build && docker run -it --rm -v builder-workspace:/workspace -v pentesting-config:/root/.pentesting -e OPENAI_API_KEY -e OPENAI_BASE_URL -e OPENAI_MODEL -e OPENAI_MAX_TOKENS agnusdei1207/pentesting:latest",
     "docker:build": "(docker image inspect agnusdei1207/pentesting-build-base:1.96 >/dev/null 2>&1 && docker image inspect agnusdei1207/pentesting-runtime-base:26.04 >/dev/null 2>&1 || npm run docker:base:build) && docker build --build-arg APP_VERSION=$(git describe --tags --abbrev=0 2>/dev/null || echo dev) -t agnusdei1207/pentesting:latest .",
     "docker:base:build": "docker build -t agnusdei1207/pentesting-build-base:1.96 -f docker/build-base.Dockerfile . && docker build -t agnusdei1207/pentesting-runtime-base:26.04 -f docker/runtime-base.Dockerfile .",
     "release:patch": "./scripts/release-all.sh patch",