projscan 4.3.1 → 4.5.0

package/docs/PLUGIN-GALLERY.md

@@ -20,16 +20,19 @@ Flags TypeScript files under a `legacy` path so teams can keep local migration
 rules close to the repo.
 
 Files:
+
 - `docs/examples/plugins/policy.projscan-plugin.json`
 - `docs/examples/plugins/policy.mjs`
 
 ### `security-radar`
 
 Flags common local security review triggers:
+
 - committed `.env` style files
 - package scripts that pipe `curl` or `wget` output into a shell
 
 Files:
+
 - `docs/examples/plugins/security-radar.projscan-plugin.json`
 - `docs/examples/plugins/security-radar.mjs`
 
@@ -38,15 +41,16 @@ Files:
 Demonstrates analyzer access to the optional graph/dataflow context. It reads the semantic graph and dataflow report through `context.getSemanticGraph()` and `context.getDataflow()` and emits a compact architecture summary issue. The test result marks `context.requested: true` and lists `semanticGraph` plus `dataflow` capabilities.
 
 Files:
+
 - `docs/examples/plugins/graph-context.projscan-plugin.json`
 - `docs/examples/plugins/graph-context.mjs`
 
-
 ### `api-route-ownership`
 
 Flags API route files that are not covered by CODEOWNERS-style routing, so PRs that change externally visible routes get a clear team owner.
 
 Files:
+
 - `docs/examples/plugins/api-route-ownership.projscan-plugin.json`
 - `docs/examples/plugins/api-route-ownership.mjs`
 
@@ -55,6 +59,7 @@ Files:
 Highlights auth, crypto, secrets, payment, middleware, and environment-related paths for explicit security-conscious review.
 
 Files:
+
 - `docs/examples/plugins/security-sensitive-files.projscan-plugin.json`
 - `docs/examples/plugins/security-sensitive-files.mjs`
 
@@ -63,6 +68,7 @@ Files:
 Flags package source files that reach across monorepo boundaries with deep relative imports instead of package entrypoints or declared workspace dependencies.
 
 Files:
+
 - `docs/examples/plugins/monorepo-boundary.projscan-plugin.json`
 - `docs/examples/plugins/monorepo-boundary.mjs`
 
@@ -73,6 +79,7 @@ Files:
 Renders `doctor`, `analyze`, and `ci` output in a compact team health voice.
 
 Files:
+
 - `docs/examples/plugins/team-radar.projscan-plugin.json`
 - `docs/examples/plugins/team-radar.mjs`
 
@@ -82,6 +89,7 @@ Renders `doctor`, `analyze`, and `ci` output as a release approval summary with
 score, blocking issue count, warnings, and a continue/hold decision.
 
 Files:
+
 - `docs/examples/plugins/release-readiness.projscan-plugin.json`
 - `docs/examples/plugins/release-readiness.mjs`
 

package/docs/ROADMAP.md

@@ -1,6 +1,6 @@
 # ProjScan Roadmap
 
-Last reviewed 2026-06-05.
+Last reviewed 2026-06-16.
 
 ---
 
@@ -18,19 +18,19 @@ projscan 1.0 shipped 2026-05-04. The stability contract is in force: MCP tool na
 
 Three forces define the next 12 months for projscan:
 
-1. **MCP is the de-facto standard.** The ecosystem has 10,000+ public servers; Claude Code, Cursor, Continue, Windsurf, and Codex all consume MCP. The protocol war is over; the value migrates to the *quality* of individual servers. Code-intelligence is one of the highest-value categories.
-2. **Multi-agent orchestration is the dominant 2026 pattern.** Claude Agent Teams, swarms, sub-agents. The new pain point is *coordination*: agents have separate context windows and need a shared source-of-truth about the codebase. projscan's graph + cache + budget-aware tools are uniquely positioned to be that shared substrate.
-3. **Context-window cost compounds.** Token spend per turn is no longer the bottleneck — it's the *accumulated* cost of carrying tool results, AST excerpts, and prior turns through every inference call. Agents that retrieve narrowly and budget aggressively win. projscan's `max_tokens`-aware response shaping, cursor pagination, and per-function chunking are exactly the primitives this trend rewards.
+1. **MCP is the de-facto standard.** The ecosystem has 10,000+ public servers; Claude Code, Cursor, Continue, Windsurf, and Codex all consume MCP. The protocol war is over; the value migrates to the _quality_ of individual servers. Code-intelligence is one of the highest-value categories.
+2. **Multi-agent orchestration is the dominant 2026 pattern.** Claude Agent Teams, swarms, sub-agents. The new pain point is _coordination_: agents have separate context windows and need a shared source-of-truth about the codebase. projscan's graph + cache + budget-aware tools are uniquely positioned to be that shared substrate.
+3. **Context-window cost compounds.** Token spend per turn is no longer the bottleneck — it's the _accumulated_ cost of carrying tool results, AST excerpts, and prior turns through every inference call. Agents that retrieve narrowly and budget aggressively win. projscan's `max_tokens`-aware response shaping, cursor pagination, and per-function chunking are exactly the primitives this trend rewards.
 
 ## The competitive picture
 
-| Tool | Position | What they do well | What we beat them on |
-|---|---|---|---|
-| **Code Pathfinder** | Direct competitor (MCP code-intel) | Deep static analysis: AST + CFG + DFG, dataflow tracking, security focus. | Language coverage (9 vs 1: Python). Composed agent tools (review / fix-suggest / impact / watch). Health signals (churn × CC, hotspots). Monorepo workspace awareness. |
-| **Sourcegraph Cody / Amp** | Enterprise paid tier | Cross-repo indexing at org scale. Polished editor integrations. | Fully offline. Open source. No SaaS dependency. Free for everyone. |
-| **Continue.dev** | Configurable RAG + MCP client | Highly extensible context providers. Local-first. | We're a *server*, not a client; we feed Continue (and every other MCP client). Different category. |
-| **Aider** | Terminal-native pair programmer | Tight Git integration, conversational refactor flow. | Different category — we're not a coding agent; we're what coding agents stand on. |
-| **GitHub MCP server** | Adjacent (repo metadata, not code intel) | Issues / PRs / Actions surface. | We do code structure; they do collaboration metadata. Complementary, not competing. |
+| Tool                       | Position                                 | What they do well                                                         | What we beat them on                                                                                                                                                   |
+| -------------------------- | ---------------------------------------- | ------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| **Code Pathfinder**        | Direct competitor (MCP code-intel)       | Deep static analysis: AST + CFG + DFG, dataflow tracking, security focus. | Language coverage (9 vs 1: Python). Composed agent tools (review / fix-suggest / impact / watch). Health signals (churn × CC, hotspots). Monorepo workspace awareness. |
+| **Sourcegraph Cody / Amp** | Enterprise paid tier                     | Cross-repo indexing at org scale. Polished editor integrations.           | Fully offline. Open source. No SaaS dependency. Free for everyone.                                                                                                     |
+| **Continue.dev**           | Configurable RAG + MCP client            | Highly extensible context providers. Local-first.                         | We're a _server_, not a client; we feed Continue (and every other MCP client). Different category.                                                                     |
+| **Aider**                  | Terminal-native pair programmer          | Tight Git integration, conversational refactor flow.                      | Different category — we're not a coding agent; we're what coding agents stand on.                                                                                      |
+| **GitHub MCP server**      | Adjacent (repo metadata, not code intel) | Issues / PRs / Actions surface.                                           | We do code structure; they do collaboration metadata. Complementary, not competing.                                                                                    |
 
 **Where we're vulnerable:** Code Pathfinder has deeper analysis (CFG, DFG) and a security-finding focus. If they ship a JavaScript or TypeScript adapter, our breadth lead narrows.
 
@@ -41,12 +41,14 @@ Three forces define the next 12 months for projscan:
 Four plays, in order:
 
 1. **Defend the lead** — close the obvious gaps so users picking an MCP server for code intel have one less reason to go elsewhere. ✅ Largely complete (1.1–1.3).
-2. **Lean into multi-agent** — make projscan the *shared substrate* for agent swarms. This is where the market is moving and where our context-budget design pays off. ✅ Largely shipped (1.4 Session, 1.5 Budgeted by default + Project Memory).
+2. **Lean into multi-agent** — make projscan the _shared substrate_ for agent swarms. This is where the market is moving and where our context-budget design pays off. ✅ Largely shipped (1.4 Session, 1.5 Budgeted by default + Project Memory).
 3. **Become the operator, not the advisor** — stop suggesting and start acting (cross-repo, apply, security gate). ✅ Shipped in the 1.6 arc.
 4. **Expand the moat** — depth where it matters (CFG / dataflow on hot paths, more languages, sub-file embeddings, cost analytics, live PR review, plugin extensibility). Not everywhere; we're not trying to be Cody. ✅ The 1.7 → 2.0 arc turns this into a platform contract.
-5. **Coordinate the swarm** — the Swarm Coordination arc. Plays 1–4 made projscan the best *single-agent* code-intelligence server; the market has moved to multi-agent orchestration, where the unsolved pain is concurrent-change arbitration across parallel agents. Turn the graph + impact + session primitives into a local-first coordination layer (collision detection, claims/leases, merge-risk preflight, intent router, one-call coordinate). ✅ Shipped additively in 3.6.0. The remaining piece — consolidating the tool surface agents pay for — is breaking, so it's reserved for **4.0**.
+5. **Coordinate the swarm** — collision detection, claims/leases, merge-risk preflight, intent routing, one-call coordination, and live coordinate watch shipped across the 3.6 through 3.7 arc, with the 4.0 tool-surface consolidation now complete. The next work is evidence: prove which commands agents reach for in real multi-worktree sessions, then deepen only the paths that prevent integration failures.
+6. **Make agent proof release-ready** — 4.1 through 4.5 turned Mission Control into a goal → mission → proof → review harness and packaged the post-4.4 implementation train: current planning surfaces, adoption examples, precise framework dataflow, scoped/redacted evidence exports, Python upgrade previews, and hotspot maintainability cleanup.
+
+We are _not_ trying to be:
 
-We are *not* trying to be:
 - A coding agent (we're what agents call into).
 - A SaaS / dashboard product.
 - A general-purpose static analyzer competing with SonarQube / Semgrep / Snyk.
@@ -54,29 +56,61 @@ We are *not* trying to be:
 
 ## Now / Next / Later
 
-### Now — Validate the Swarm Coordination arc; prepare the 4.0 surface break
+### Now — Post-4.5 Validation
+
+4.5.0 "Review-Ready Intelligence Train" packages the post-4.4 implementation train. The next work is validation and selective hardening from real use, not another broad feature push.
+
+The active validation lines are:
+
+- **Swarm coordination evidence.** Validate how real agents use `collisions`, `claim`, `merge-risk`, `coordinate`, and `coordinate --watch`; deepen only the coordination paths that prevent integration failures.
+- **Evidence export adoption.** Prove scoped/redacted report controls work for partner, security, and release-review handoffs without leaking unnecessary repo structure.
+- **Python upgrade coverage.** Extend lockfile support only after Poetry and pinned-requirement evidence prove useful in real repos.
+- **Framework dataflow precision.** Add more framework patterns only when each has a narrow request source, sink, and false-positive fixture.
+- **Hotspot maintainability.** Continue extracting and covering high-churn start/review/type surfaces when they show concrete review or defect risk.
+
+Strictly **local-first** throughout: same-repo / same-machine evidence, no daemon, no cloud, no hidden network calls, no new telemetry, and no secret-value reads.
+
+Success signals: teams copy the adoption examples into real reviews, scoped/redacted artifacts are accepted by reviewers, Python upgrade previews identify useful local evidence, dataflow additions stay quiet on lookalikes, and release bug-hunts remain free of concrete defects.
+
+### Recently Completed — 4.5.0 (2026)
+
+**4.5.0 "Review-Ready Intelligence Train"** shipped the post-4.4 implementation train:
+
+- Roadmap and release-train planning now default to the current post-4.4 product lines instead of stale shipped work.
+- Adoption examples cover agent orchestration, package ownership, custom policy plugins, swarm coordination, and scoped evidence exports.
+- `analyze`, `doctor`, and `ci` can scope and redact shareable evidence with direct flags or named `reportPolicies` presets.
+- `projscan upgrade` and MCP `projscan_upgrade` support offline Python previews from manifests, Poetry lockfiles, pinned requirements, and Python importers.
+- Dataflow detects narrow Fastify and Koa request-source patterns while suppressing lookalike helpers and Koa response-body writes.
+- Start next-action assembly and taint function identity were tightened during release readiness cleanup.
 
-The **Swarm Coordination arc shipped in 3.6.0** (see Recently Completed). It was additive — five new tools, nothing removed — so it shipped as a minor release, not a major one. ("4.x" was an earlier label for the *theme*; the version is 3.6.0. The name **4.0** is reserved strictly for the one *breaking* change below.)
+### Recently Completed — 4.4.0 (2026)
 
-What's now:
+**4.4.0 "Agent Release Harness"** turned Mission Control into a release-ready agent harness:
 
-- **Validate it in real swarm usage.** The arc is built on an unvalidated bet that concurrent-change arbitration is the pain. Before deepening it, find out which of `collision` / `claim` / `merge-risk` / `coordinate` agents actually reach for, and harden from there (transitive collision recall, live `--watch` coordination, integration into `preflight` / `agent_brief`).
-- **`4.0` — tool-surface consolidation (the first breaking release since 1.0).** In progress on the `next` branch (publish held for a real deprecation window). 4.0 removes the two tools deprecated in 3.8.0 — `projscan_explain` (use `projscan_file`) and `projscan_graph` (use `projscan_semantic_graph`, which gains a targeted `query` mode that subsumes it) — taking the surface from 47 → 45. Both have drop-in replacements (see [MIGRATION-4.0.md](MIGRATION-4.0.md)). The *broader* consolidation (routing the long tail behind `projscan_route`) is deliberately deferred until real usage signal justifies which tools to fold — same deprecate-before-remove discipline.
+- Repo-local AgentLoopKit and AgentFlight harness commands are surfaced as proof hints when harness files exist.
+- Product-planning intents route to verifiable bug-hunt/action planning instead of generic orientation.
+- Bug-hunt, release-train, evidence-pack, and review wording distinguish concrete fix targets from manual release sign-off actions.
+- Public type contracts are split into focused modules with a dedicated `typecheck:public-types` gate.
+- Same-SHA dirty-worktree review and directory-only verification guidance were fixed.
+- The dev dependency chain cleared the release audit gate without adding runtime dependencies.
 
-Strictly **local-first** throughout: same-repo / same-machine swarms via the shared store, never a daemon, cloud, or cross-machine server (that would be a SaaS non-goal).
+### Recently Completed — 4.0.0 through 4.3.1 (2026)
 
-Success signals: collisions prevented pre-merge, integration-failure-rate reduction, tokens saved per turn via the router, first external swarm adopter.
+- **4.0.0 "Surface Consolidation"** removed the deprecated MCP tools `projscan_explain` and `projscan_graph` after a documented deprecation cycle. CLI commands were not removed. `projscan_file` and `projscan_semantic_graph` query mode are the replacements.
+- **4.1.0 through 4.2.0 "Mission Control Handoffs"** added execution plans, cursors, runbooks, task cards, review gates, shortcut commands, and saved mission bundles.
+- **4.3.0 "Mission Outcome Loop"** added `projscan start --mission <dir>` and Mission Proof outcome summaries.
+- **4.3.1 "Mission Proof Polish"** added Markdown proof reports, saved proof output, newest/all-bundle selection, attention filters, one-line CI summaries, and reproducible demo media.
 
-### Recently Completed — 3.6.0 (2026)
+### Recently Completed — 3.6.0 and 3.7.x Coordination (2026)
 
-**3.6.0 "Swarm Coordination"** turned projscan into the local-first coordination substrate for parallel agents working one repo across git worktrees:
+The **Swarm Coordination arc** turned projscan into the local-first coordination substrate for parallel agents working one repo across git worktrees:
 
-- `projscan collisions` / `projscan_collision` — same-file and dependency overlaps across in-flight worktrees, surfaced before the branches merge (reuses the import graph for blast radius).
-- `projscan claim` / `projscan_claim` — advisory claims/leases over files, dirs, or symbols, shared across worktrees, with `--ttl` expiry (so a crashed agent's claim auto-expires), contention warnings, and `prune`.
-- `projscan merge-risk` / `projscan_merge_risk` — safe integration order (merge the least-entangled branch first) plus conflict hotspots.
-- `projscan route` / `projscan_route` — map a stated goal to the right tool (additive discovery entry over the surface; deterministic, no LLM).
-- `projscan coordinate` / `projscan_coordinate` — one-call read folding it all into a `clear` / `caution` / `conflicted` readiness verdict.
-- Also: semantic search degrades to BM25 when the embedding model can't load instead of crashing. 41 → 47 MCP tools.
+- `projscan collisions` / `projscan_collision` — same-file and dependency overlaps across in-flight worktrees, surfaced before branches merge.
+- `projscan claim` / `projscan_claim` — advisory claims/leases over files, dirs, or symbols, shared across worktrees, with `--ttl` expiry, contention warnings, and `prune`.
+- `projscan merge-risk` / `projscan_merge_risk` — safe integration order plus conflict hotspots.
+- `projscan route` / `projscan_route` — deterministic goal-to-tool routing.
+- `projscan coordinate` / `projscan_coordinate` — one-call readiness verdict over collisions, claims, and merge risk.
+- `projscan coordinate --watch` / `projscan_coordinate_watch` — local polling with MCP coordination-change notifications.
 
 ### Recently Completed — 3.5.0 (2026)
 
@@ -132,11 +166,11 @@ Success signals: collisions prevented pre-merge, integration-failure-rate reduct
 
 ### Later
 
-Later work should expand the moat after 3.4.0 repo-understanding output is verified in real engineering workflows:
+Later work should deepen proven surfaces rather than add broad categories:
 
-- Broaden framework dataflow precision from narrow, tested source patterns rather than broad source-name matching.
-- Add adoption examples from real agent orchestration, package ownership, and custom policy plugin workflows.
-- Explore stronger report-export controls for teams that want path redaction or scoped evidence artifacts.
+- Add stronger sub-file and symbol-level coordination once real swarm examples show which conflicts matter.
+- Extend Python upgrade intelligence toward broader lockfile formats after Poetry and pinned-requirement evidence are proven useful.
+- Add more framework dataflow patterns only when each has a clear request source, sink, and false-positive fixture.
 
 ## Non-goals
 
@@ -149,7 +183,7 @@ Later work should expand the moat after 3.4.0 repo-understanding output is verif
 ## Risks
 
 - **Code Pathfinder catches up on languages.** They're 1-language today (Python) but the AST + CFG infrastructure is solid. If they ship a JS/TS adapter, our breadth lead narrows. Mitigation: keep adding languages on the cadence; deepen agent-native composition.
-- **Multi-agent orchestration matures faster than we can ship Session.** If Claude Agent Teams becomes the default and ships its own shared-state primitive, our 1.4 bet weakens. Mitigation: design Session as a *complement* to Agent Teams rather than a replacement.
+- **Multi-agent orchestration matures faster than we can ship Session.** If Claude Agent Teams becomes the default and ships its own shared-state primitive, our 1.4 bet weakens. Mitigation: design Session as a _complement_ to Agent Teams rather than a replacement.
 - **Context-cost trend reverses.** If models get cheaper and context windows grow, our budget-aware design becomes table stakes rather than a differentiator. Mitigation: that's a good problem to have; the underlying primitives still work.
 
 ## How to influence this roadmap
@@ -165,55 +199,55 @@ If you've adopted projscan and want something specific:
 
 For the full release notes, see [CHANGELOG.md](../CHANGELOG.md).
 
-| Version | Theme | Headline |
-|---|---|---|
-| **3.6.0** (2026-06-05) | Swarm Coordination | Local-first coordination for parallel agents across git worktrees: `projscan_collision`, `projscan_claim` (leased), `projscan_merge_risk`, `projscan_route`, `projscan_coordinate`; graceful embedding degradation; 41 → 47 tools |
-| **3.5.0** (2026-06-04) | Plugin Trust | `projscan fix` installs with `--ignore-scripts` (no lifecycle-script RCE); local plugins gated by trust-on-first-use (`projscan plugin trust`), plus a hardened, model-degrading embedding path |
-| **3.4.0** (2026-06-04) | Repo Understanding | `projscan understand` / `projscan_understand` with cited repo, flow, contract, change-readiness, and verification maps for working engineers |
-| **3.3.0** (2026-06-03) | Roadmap Evidence Polish | Adoption proof gates, reviewer decision evidence, first-ten-minutes/start coordination hints, Hono request-source precision, plugin trust guidance, generated PR-comment validation, and evidence helper extraction |
-| **3.2.0** (2026-06-03) | Roadmap Train | Canonical 3.2-3.9 release train surfaced in release planning and roadmap docs |
-| **3.1.0** (2026-06-02) | Trust Boundary Hardening | Privacy-check trust report, Git-visible scan boundary, path-only `.env` defaults, offline mode, session/worktree risk split, and fast trust smoke gate |
-| **3.0.8** (2026-06-01) | Legal and Trust Hardening | Public legal/trust documents, vulnerability reporting, contribution provenance, and canonical icon packaging |
-| **3.0.7** (2026-05-31) | Trial Adoption Report | Adoption trial verdict, structured reviewer feedback capture, measured market-validation gates, and refreshed adoption docs |
-| **3.0.6** (2026-05-31) | Market Validation Loop | Feedback-backed dogfood evidence, minutes-saved/prevented-edit tracking, false-positive reporting, Baseframe Labs brand surfaces, and security disclosure assets |
-| **3.0.5** (2026-05-28) | Proof of Usefulness | End-to-end adoption harness, five-scenario PR comment benchmarks, fix-first output, richer baseline trend memory, Express/Next dataflow precision, and scale-risk calibration |
-| **3.0.4** (2026-05-28) | Team Adoption Loop | Team bootstrap, MCP setup doctor, validated PR evidence comments, baseline trend memory, owner routing, practical plugins, and trust calibration |
-| **3.0.3** (2026-05-27) | Agent Review Precision | Package-scoped review verdicts, receiver-sensitive route request sources, generated-code review/dataflow filtering, package owner fallback, and v5 GitHub Actions |
-| **3.0.2** (2026-05-27) | Agent Graph Readiness | Graph corpus release gates, custom dataflow visibility, remote tag integrity, CODEOWNERS impact ownership, and 3.x release-train planning |
-| **3.0.1** (2026-05-26) | Graph Operations Platform | Graph-backed review/workplan/brief evidence, cross-repo boundary impact, plugin graph context, golden graph corpus, and hardened dataflow precision |
-| **3.0.0** (2026-05-23) | Deep Graph Platform | Stable v3 semantic graph, dataflow risk engine, bridge-helper review blocks, 39-tool MCP surface, and public graph/dataflow APIs |
-| **2.9.0** (2026-05-23) | Adoption Layer | MCP client config snippets, workflow recipes, first-run diagnostics, adoption MCP tool, plugin gallery, and console guidance polish |
-| **2.8.0** (2026-05-22) | Agent Mission Control | Workplans, bug-hunt queues, release readiness, evidence packs, regression plans, agent briefs, and quality scorecards |
-| **2.0.0** (2026-05-18) | Plugin Platform | Stable local analyzer/reporter plugin contract, manifest schema and tested examples, CLI JSON `schemaVersion: 2`, extensible `LanguageId`, and removal of deprecated regex import/export helpers |
-| **1.11.0** (2026-05-18) | Reporter Plugins | Reporter plugin preview for CLI output (`--reporter` on `doctor`, `analyze`, and `ci`), reporter manifest validation through `projscan_plugin`, and refreshed README media with a macOS-style terminal demo |
-| **1.10.0** (2026-05-13) | RC for 2.0 | Analyzer plugin API preview behind PROJSCAN_PLUGINS_PREVIEW flag (`projscan_plugin` MCP tool, `projscan plugin` CLI, `.projscan-plugins/*.projscan-plugin.json` schema); live cost-summary streaming with `notifications/projscan/cost_delta`; five 1.9-deferred fixes (applyFix rollback dir handling, incrementalUpdateGraph context staleness, changedFiles maxBuffer surfacing, taint per-step frontier cap, watcher.close mid-flush) |
-| **1.9.0** (2026-05-12) | Intent + Polish | Intent-grounded review: free-text PR description → per-finding expected / unexpected / out-of-scope labels (no LLM); Project Memory loop #4 (per-rule severity drift, cry-wolf / noisy / stable); review_watch signature deepening with structured `delta` payload; macOS CI leg |
-| **1.8.1** (2026-05-08) | Docs patch | README setup snippets for Codex CLI + Gemini CLI |
-| **1.8.0** (2026-05-08) | Resilience + Live | Swift adapter (11 languages); long-running PR watch (`projscan_review_watch`) with `notifications/projscan/pr_changed`; atomic session save; taint truncation reporting + MAX_DEPTH 8 → 12; embeddings LRU; templated C++ qualified-id; 7 fixes from a three-way multi-agent bug hunt |
-| **1.7.0** (2026-05-07) | Reach + Visibility | Kotlin and C++ adapters (10 languages); per-rule confidence in Project Memory (loop #3); aggregate cost analytics (`projscan_cost_summary`); 6 fixes from a four-way multi-agent bug hunt |
-| **1.6.0** (2026-05-06) | Operator | Cross-repo workspace + intelligence (`projscan_workspace_graph`); mechanical apply layer with rollback (`projscan_apply_fix`, six templates); source-to-sink taint analysis (`projscan_taint`) wired into review as a hard block on new flows |
-| **1.5.0** (2026-05-05) | Budgeted by default | `_cost` sidecar on every result; adaptive `projscan_review` with full / summary / verdict-only tiers |
-| **1.4.0** (2026-05-05) | Session | Durable cross-invocation session: `projscan_session` MCP tool, auto-touched files, event log |
-| **1.3.0** (2026-05-05) | Push, Don't Poll | MCP `notifications/file_changed` push and registry-aware upgrade preview |
-| **1.2.1** (2026-05-05) | Animated docs | Animated GIFs replace static command screenshots |
-| **1.2.0** (2026-05-05) | Reporter Parity | PHP and C# adapters, HTML reporters, per-function fan-out |
-| **1.1.1** (2026-05-04) | Dogfood patch | Tree-sitter false-positive fix |
-| **1.1.0** (2026-05-04) | On the Map | Rust adapter and fix-suggest templates for `eslint-*` and `python-type-error-*` |
-| **1.0.0** (2026-05-04) | Stable | Public no-break commitment release |
-| **0.17.0** (2026-05-02) | RC + Docs | Documentation reorganized around the agent journey |
-| **0.16.0** (2026-04-30) | Live | `projscan watch` CLI and HTML report export |
-| **0.15.0** (2026-04-27) | Reach | `projscan_impact` blast-radius tool, per-function fan-in, sub-file embeddings |
-| **0.14.0** (2026-04-26) | Agent Fix Loop | `projscan_fix_suggest` and `projscan_explain_issue` |
-| **0.13.0** (2026-04-26) | Agent Review | `projscan_review` one-call PR review and per-function cyclomatic complexity |
-| **0.12.0** (2026-04-25) | — | Java and Ruby adapters, workspace-aware `outdated` and unused-dep |
-| **0.11.0** (2026-04-25) | — | AST cyclomatic complexity, `projscan_coupling`, `projscan_pr_diff`, monorepo workspace detection, Go adapter |
-| **0.10.0** (2026-04-24) | Beyond JS | Python as a first-class language; `LanguageAdapter` interface |
-| **0.9.0–0.9.2** (2026-04-20) | True Semantic Search | Optional `@xenova/transformers` peer; security patch for path traversal |
-| **0.8.0** (2026-04-20) | Streaming & Pagination | MCP protocol 2025-03-26, cursor pagination, progress notifications |
-| **0.7.0** (2026-04-20) | Smart Search | BM25-ranked content + symbol + path search |
-| **0.6.0** (2026-04-20) | Agent-First | Real AST parsing, code graph primitive, incremental cache, MCP token budgeter |
-| **0.5.0** (2026-04-20) | Deeper Signal | `projscan coverage`, dead-code analyzer |
-| **0.4.0** (2026-04-20) | Dependency Health | `projscan outdated` / `audit` / `upgrade`, unused-dependency analyzer |
-| **0.3.0–0.3.1** (2026-04-20) | — | SARIF output, `--changed-only`, `.projscanrc` config, GitHub Action |
-| **0.2.0** (2026-04-19) | — | `projscan hotspots`, `projscan mcp` |
-| **0.1.x** (2026-03-11) | — | Initial release: analyze, doctor, fix, explain, diagram, structure, dependencies, badge |
+| Version                      | Theme                     | Headline                                                                                                                                                                                                                                                                                                                                                                                                                                  |
+| ---------------------------- | ------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| **3.6.0** (2026-06-05)       | Swarm Coordination        | Local-first coordination for parallel agents across git worktrees: `projscan_collision`, `projscan_claim` (leased), `projscan_merge_risk`, `projscan_route`, `projscan_coordinate`; graceful embedding degradation; 41 → 47 tools                                                                                                                                                                                                         |
+| **3.5.0** (2026-06-04)       | Plugin Trust              | `projscan fix` installs with `--ignore-scripts` (no lifecycle-script RCE); local plugins gated by trust-on-first-use (`projscan plugin trust`), plus a hardened, model-degrading embedding path                                                                                                                                                                                                                                           |
+| **3.4.0** (2026-06-04)       | Repo Understanding        | `projscan understand` / `projscan_understand` with cited repo, flow, contract, change-readiness, and verification maps for working engineers                                                                                                                                                                                                                                                                                              |
+| **3.3.0** (2026-06-03)       | Roadmap Evidence Polish   | Adoption proof gates, reviewer decision evidence, first-ten-minutes/start coordination hints, Hono request-source precision, plugin trust guidance, generated PR-comment validation, and evidence helper extraction                                                                                                                                                                                                                       |
+| **3.2.0** (2026-06-03)       | Roadmap Train             | Canonical 3.2-3.9 release train surfaced in release planning and roadmap docs                                                                                                                                                                                                                                                                                                                                                             |
+| **3.1.0** (2026-06-02)       | Trust Boundary Hardening  | Privacy-check trust report, Git-visible scan boundary, path-only `.env` defaults, offline mode, session/worktree risk split, and fast trust smoke gate                                                                                                                                                                                                                                                                                    |
+| **3.0.8** (2026-06-01)       | Legal and Trust Hardening | Public legal/trust documents, vulnerability reporting, contribution provenance, and canonical icon packaging                                                                                                                                                                                                                                                                                                                              |
+| **3.0.7** (2026-05-31)       | Trial Adoption Report     | Adoption trial verdict, structured reviewer feedback capture, measured market-validation gates, and refreshed adoption docs                                                                                                                                                                                                                                                                                                               |
+| **3.0.6** (2026-05-31)       | Market Validation Loop    | Feedback-backed dogfood evidence, minutes-saved/prevented-edit tracking, false-positive reporting, Baseframe Labs brand surfaces, and security disclosure assets                                                                                                                                                                                                                                                                          |
+| **3.0.5** (2026-05-28)       | Proof of Usefulness       | End-to-end adoption harness, five-scenario PR comment benchmarks, fix-first output, richer baseline trend memory, Express/Next dataflow precision, and scale-risk calibration                                                                                                                                                                                                                                                             |
+| **3.0.4** (2026-05-28)       | Team Adoption Loop        | Team bootstrap, MCP setup doctor, validated PR evidence comments, baseline trend memory, owner routing, practical plugins, and trust calibration                                                                                                                                                                                                                                                                                          |
+| **3.0.3** (2026-05-27)       | Agent Review Precision    | Package-scoped review verdicts, receiver-sensitive route request sources, generated-code review/dataflow filtering, package owner fallback, and v5 GitHub Actions                                                                                                                                                                                                                                                                         |
+| **3.0.2** (2026-05-27)       | Agent Graph Readiness     | Graph corpus release gates, custom dataflow visibility, remote tag integrity, CODEOWNERS impact ownership, and 3.x release-train planning                                                                                                                                                                                                                                                                                                 |
+| **3.0.1** (2026-05-26)       | Graph Operations Platform | Graph-backed review/workplan/brief evidence, cross-repo boundary impact, plugin graph context, golden graph corpus, and hardened dataflow precision                                                                                                                                                                                                                                                                                       |
+| **3.0.0** (2026-05-23)       | Deep Graph Platform       | Stable v3 semantic graph, dataflow risk engine, bridge-helper review blocks, 39-tool MCP surface, and public graph/dataflow APIs                                                                                                                                                                                                                                                                                                          |
+| **2.9.0** (2026-05-23)       | Adoption Layer            | MCP client config snippets, workflow recipes, first-run diagnostics, adoption MCP tool, plugin gallery, and console guidance polish                                                                                                                                                                                                                                                                                                       |
+| **2.8.0** (2026-05-22)       | Agent Mission Control     | Workplans, bug-hunt queues, release readiness, evidence packs, regression plans, agent briefs, and quality scorecards                                                                                                                                                                                                                                                                                                                     |
+| **2.0.0** (2026-05-18)       | Plugin Platform           | Stable local analyzer/reporter plugin contract, manifest schema and tested examples, CLI JSON `schemaVersion: 2`, extensible `LanguageId`, and removal of deprecated regex import/export helpers                                                                                                                                                                                                                                          |
+| **1.11.0** (2026-05-18)      | Reporter Plugins          | Reporter plugin preview for CLI output (`--reporter` on `doctor`, `analyze`, and `ci`), reporter manifest validation through `projscan_plugin`, and refreshed README media with a macOS-style terminal demo                                                                                                                                                                                                                               |
+| **1.10.0** (2026-05-13)      | RC for 2.0                | Analyzer plugin API preview behind PROJSCAN_PLUGINS_PREVIEW flag (`projscan_plugin` MCP tool, `projscan plugin` CLI, `.projscan-plugins/*.projscan-plugin.json` schema); live cost-summary streaming with `notifications/projscan/cost_delta`; five 1.9-deferred fixes (applyFix rollback dir handling, incrementalUpdateGraph context staleness, changedFiles maxBuffer surfacing, taint per-step frontier cap, watcher.close mid-flush) |
+| **1.9.0** (2026-05-12)       | Intent + Polish           | Intent-grounded review: free-text PR description → per-finding expected / unexpected / out-of-scope labels (no LLM); Project Memory loop #4 (per-rule severity drift, cry-wolf / noisy / stable); review_watch signature deepening with structured `delta` payload; macOS CI leg                                                                                                                                                          |
+| **1.8.1** (2026-05-08)       | Docs patch                | README setup snippets for Codex CLI + Gemini CLI                                                                                                                                                                                                                                                                                                                                                                                          |
+| **1.8.0** (2026-05-08)       | Resilience + Live         | Swift adapter (11 languages); long-running PR watch (`projscan_review_watch`) with `notifications/projscan/pr_changed`; atomic session save; taint truncation reporting + MAX_DEPTH 8 → 12; embeddings LRU; templated C++ qualified-id; 7 fixes from a three-way multi-agent bug hunt                                                                                                                                                     |
+| **1.7.0** (2026-05-07)       | Reach + Visibility        | Kotlin and C++ adapters (10 languages); per-rule confidence in Project Memory (loop #3); aggregate cost analytics (`projscan_cost_summary`); 6 fixes from a four-way multi-agent bug hunt                                                                                                                                                                                                                                                 |
+| **1.6.0** (2026-05-06)       | Operator                  | Cross-repo workspace + intelligence (`projscan_workspace_graph`); mechanical apply layer with rollback (`projscan_apply_fix`, six templates); source-to-sink taint analysis (`projscan_taint`) wired into review as a hard block on new flows                                                                                                                                                                                             |
+| **1.5.0** (2026-05-05)       | Budgeted by default       | `_cost` sidecar on every result; adaptive `projscan_review` with full / summary / verdict-only tiers                                                                                                                                                                                                                                                                                                                                      |
+| **1.4.0** (2026-05-05)       | Session                   | Durable cross-invocation session: `projscan_session` MCP tool, auto-touched files, event log                                                                                                                                                                                                                                                                                                                                              |
+| **1.3.0** (2026-05-05)       | Push, Don't Poll          | MCP `notifications/file_changed` push and registry-aware upgrade preview                                                                                                                                                                                                                                                                                                                                                                  |
+| **1.2.1** (2026-05-05)       | Animated docs             | Animated GIFs replace static command screenshots                                                                                                                                                                                                                                                                                                                                                                                          |
+| **1.2.0** (2026-05-05)       | Reporter Parity           | PHP and C# adapters, HTML reporters, per-function fan-out                                                                                                                                                                                                                                                                                                                                                                                 |
+| **1.1.1** (2026-05-04)       | Dogfood patch             | Tree-sitter false-positive fix                                                                                                                                                                                                                                                                                                                                                                                                            |
+| **1.1.0** (2026-05-04)       | On the Map                | Rust adapter and fix-suggest templates for `eslint-*` and `python-type-error-*`                                                                                                                                                                                                                                                                                                                                                           |
+| **1.0.0** (2026-05-04)       | Stable                    | Public no-break commitment release                                                                                                                                                                                                                                                                                                                                                                                                        |
+| **0.17.0** (2026-05-02)      | RC + Docs                 | Documentation reorganized around the agent journey                                                                                                                                                                                                                                                                                                                                                                                        |
+| **0.16.0** (2026-04-30)      | Live                      | `projscan watch` CLI and HTML report export                                                                                                                                                                                                                                                                                                                                                                                               |
+| **0.15.0** (2026-04-27)      | Reach                     | `projscan_impact` blast-radius tool, per-function fan-in, sub-file embeddings                                                                                                                                                                                                                                                                                                                                                             |
+| **0.14.0** (2026-04-26)      | Agent Fix Loop            | `projscan_fix_suggest` and `projscan_explain_issue`                                                                                                                                                                                                                                                                                                                                                                                       |
+| **0.13.0** (2026-04-26)      | Agent Review              | `projscan_review` one-call PR review and per-function cyclomatic complexity                                                                                                                                                                                                                                                                                                                                                               |
+| **0.12.0** (2026-04-25)      | —                         | Java and Ruby adapters, workspace-aware `outdated` and unused-dep                                                                                                                                                                                                                                                                                                                                                                         |
+| **0.11.0** (2026-04-25)      | —                         | AST cyclomatic complexity, `projscan_coupling`, `projscan_pr_diff`, monorepo workspace detection, Go adapter                                                                                                                                                                                                                                                                                                                              |
+| **0.10.0** (2026-04-24)      | Beyond JS                 | Python as a first-class language; `LanguageAdapter` interface                                                                                                                                                                                                                                                                                                                                                                             |
+| **0.9.0–0.9.2** (2026-04-20) | True Semantic Search      | Optional `@xenova/transformers` peer; security patch for path traversal                                                                                                                                                                                                                                                                                                                                                                   |
+| **0.8.0** (2026-04-20)       | Streaming & Pagination    | MCP protocol 2025-03-26, cursor pagination, progress notifications                                                                                                                                                                                                                                                                                                                                                                        |
+| **0.7.0** (2026-04-20)       | Smart Search              | BM25-ranked content + symbol + path search                                                                                                                                                                                                                                                                                                                                                                                                |
+| **0.6.0** (2026-04-20)       | Agent-First               | Real AST parsing, code graph primitive, incremental cache, MCP token budgeter                                                                                                                                                                                                                                                                                                                                                             |
+| **0.5.0** (2026-04-20)       | Deeper Signal             | `projscan coverage`, dead-code analyzer                                                                                                                                                                                                                                                                                                                                                                                                   |
+| **0.4.0** (2026-04-20)       | Dependency Health         | `projscan outdated` / `audit` / `upgrade`, unused-dependency analyzer                                                                                                                                                                                                                                                                                                                                                                     |
+| **0.3.0–0.3.1** (2026-04-20) | —                         | SARIF output, `--changed-only`, `.projscanrc` config, GitHub Action                                                                                                                                                                                                                                                                                                                                                                       |
+| **0.2.0** (2026-04-19)       | —                         | `projscan hotspots`, `projscan mcp`                                                                                                                                                                                                                                                                                                                                                                                                       |
+| **0.1.x** (2026-03-11)       | —                         | Initial release: analyze, doctor, fix, explain, diagram, structure, dependencies, badge                                                                                                                                                                                                                                                                                                                                                   |