Skip to content

TODO.md

Latest commit

 

History

History
591 lines (505 loc) · 33.9 KB

TODO.md

File metadata and controls

591 lines (505 loc) · 33.9 KB
Raw

TODO — Forgeplan

Roadmap: see docs/ROADMAP.md for full gap analysis by category (Architecture 85%, UX 70%, Performance 80%, Distribution 65%, Docs 60%, Integrations 55%).

Current: v0.34.0 (released 2026-08-17 — Cargo.toml workspace = 0.34.0)

Recent releases (authoritative: git tag --sort=-v:refname | head + Cargo.toml):

  • v0.34.0 (released 2026-08-17) — artifact integrity: R_eff scores current evidence only (ADR-020, breaking) + git-delta provenance for code-claiming Evidence (PRD-082/#360) + identity/collision hardening (update --status gated, reindex duplicate-id detection, unbacked_displacement detector, anomalies 140 → 34). Security: RUSTSEC-2026-0204 + quinn-proto HIGH + serde_with.

  • v0.33.0 (released 2026-06-04) — MCP worktree-aware routing (PRD-078) + CRITICAL #350 @file expansion + PROB-078 refuted.

  • v0.32.1 / v0.32.0 (released 2026-05-21) — Windows binary hotfix; Epic #287 brownfield surface + PROB-074 stale-handle hardening.

  • v0.31.0 (released 2026-05-12, sync PR #285) — Wave 9 quality closure.

  • v0.30.0 (released 2026-05-06, sync PR #262) — PROB-060 slug-canonical identity marathon (PR #255..#274) + Phase 2.5 resolver wire-up.

  • v0.29.0 (released 2026-05-05, sync PR #250) — verdict aggregator typed errors (PROB-049 H-class), claude --print dispatch refactor (PROB-050 A-4..A-15), CWE-426 binary substitution closed.

  • v0.28.0 (released 2026-05-04, sync PR #223) — file-first invariant compile-enforced (PRD-073) + claude --print dispatchers (ADR-011) + canonical playbooks (release/brownfield-docs).

In-flight на dev к v0.31.0 (PRs merged after v0.30.0 tag):

  • PR #275 — PROB-060 Phase 2.5 test coverage extension — merged 2026-05-10.
  • PR #277 — PROB-063 verdict aggregator excludes advisory_phase_mismatches (Closes #276) — open, awaiting CI green at время этой записи.

Released: v0.28.0 — file-first invariant compile-enforced + claude --print + canonical playbooks

Bundles 14 merge-PR (#224..#237) since v0.27.0 (2026-04-28). Three load-bearing themes:

  1. PRD-073 file-first invariant compile-enforced (ADR-003) — LanceStore::* mutating methods stали pub(crate), file-first projection wrappers — единственная mutation surface. EVID-094 R_eff=0.80 grade A. PROB-048 deprecated.
  2. ADR-011 Phase B Wave 1 — Plugin/Agent dispatchers shell out to claude --print на real claude 2.1.126. Replaces fictional task-tool из ADR-010. EVID-093 + EVID-096 + EVID-097, R_eff=0.70 grade B. Real-E2E verified.
  3. Track 4-A8 canonical playbooks — release.yaml + brownfield-docs.yaml ship как REFERENCE templates для marketplace skill/mapping authors. audit.yaml migrated к Plugin variant + budget_usd=$5 (PROB-050 A-28 closed).

PRD-073 — file-first invariant compile-enforced

  • Phase 3a: 15 projection helpers
  • Phase 3b: 16 mutation helpers migrated к MutationResult<T>
  • Phase 3c: typed errors (R1 + R2 audit closures)
  • Phase 4: LanceStore::* mutating methods → pub(crate) lockdown
  • EVID-094 (PRD-073 closure) + EVID-095 (Phase 3c sprint closure)
  • PROB-048 deprecated as resolved

ADR-011 — Phase B Wave 1 (claude --print dispatchers)

  • PluginDispatcher + AgentDispatcher rewritten для claude --print
  • Step.budget_usd + Step.allowed_tools fields (SPEC-003 1.2)
  • R1 audit closure (4 CRITICAL + 18 HIGH/MEDIUM)
  • EVID-093 spike + EVID-096 closure measurement
  • PR 1 (NOTE-049 + EVID-097) — real-E2E на production binary, 5 invocations, ~$0.98
  • A-28 closure (audit.yaml type:agent → type:plugin) + real chain run ~$3.50

Release v0.28.0 (this sprint)

  • CHANGELOG [Unreleased] → [0.28.0] promotion
  • Cargo.toml workspace 0.27.0 → 0.28.0 + 4 internal crate refs
  • Pre-flight: cargo fmt + cargo clippy -D warnings + cargo test (1940 PASS)
  • Audit (architect + code-analyzer adversarial): 11 findings closed inline
  • Audit round 2 (security + code-analyzer на PR 1 verification): 11 findings closed
  • Dependabot triage round 3 (18 alerts unchanged from round 2)
  • EVID-098 readiness measurement (R_eff propagation to PRD-073 grade A)
  • AI docs accessibility: robots.txt + llms.txt
  • MCP tool count drift swept (63 canonical, 18 doc locations updated)
  • Push release/v0.28.0 → main (PR #238 merged 2026-05-04)
  • Tag v0.28.0 + cargo-dist + brew publish (released 2026-05-04)
  • Post-release sync PR chore/sync-main-to-dev-after-v0.28.0 (PR #223 merged 2026-04-28 — predecessor; v0.28.0 sync через cascade)

Follow-ups (deferred к PR 3 / PR 4)

  • PR 3: PROB-049 top-4 (StoreError split, # Errors rustdoc, MutationContext, let-else)
  • PR 4: PROB-050 top-5 (SPEC-003 1.2 doc, claude_print::invoke() extract, cross-file ENV lock, API surface tighten, integration test gated by CLAUDE_BIN_AVAILABLE)
  • PROB-050 A-21..A-29: real-E2E findings (discovery, exit codes already-OK, version disambiguation, budget tier, audit.yaml budget=$5, methodology hardening)

Previous: v0.27.0 — PRD-072 Phase 6 real dispatchers + init wiring + greenfield playbook

EPIC-007 Phase 6 — engine layer (v0.26.0) переходит в user-facing activation. PRD-072 / RFC-007 / ADR-010 закрывают Phase 5 deferral: 5 production Dispatcher impls (real subprocess через tokio::process + ForgeplanCore direct call), forgeplan init теперь эмитит recommendation hints, и greenfield-kickoff.yaml доступен в marketplace. 3 waves × 8 agents, ~5000 LOC, +60 unit + 5 integration tests.

PRD-072 — Phase 6 real subprocess dispatchers

  • FR-1 PluginDispatcher (claude-code-plugin subprocess, default 600s)
  • FR-2 AgentDispatcher (task-tool agent-invoke, default 300s)
  • FR-3 SkillDispatcher (in-process v1 stub — Wave 5 real registry)
  • FR-4 CommandDispatcher (security-hardened: env_clear, no shell, --yes)
  • FR-5 ForgeplanCoreDispatcher (direct internal call)
  • FR-6 commands::init::run recommendation wiring
  • FR-7 marketplace/playbooks/greenfield-kickoff.yaml canonical (7 steps)
  • FR-9 Subprocess lifecycle (tokio::process + kill_on_drop + Stdio::piped + 10 MiB cap)
  • FR-10 Security (env_clear, allow-list, no shell expansion, --yes gate)
  • AC-3..AC-7 init recommendation hints (закрывает PRD-067)
  • AC-7 + AC-8 greenfield-kickoff validate + dry-run pass
  • AC-9 regression-free (1384+ lib + 372+ integration tests PASS)
  • FR-8 Per-step timeout_seconds override (schema landed; executor wiring partial — Wave 5)
  • AC-1, AC-2 real brownfield E2E run on real c4-architecture installed plugin (Wave 5)
  • AC-10 kill -9 mid-step resumability E2E (Wave 5)

Documentation + Marketplace (Phase 6)

  • docs/operations/PLAYBOOK-AUTHORING.ru.md Subprocess lifecycle section (Wave 4)
  • docs/operations/INGEST-MAPPINGS.ru.md Workflow integration update (Wave 4)
  • marketplace/playbooks/greenfield-kickoff.yaml canonical (Wave 3)
  • CHANGELOG.md v0.27.0 section (Wave 4)
  • EVID-091 — Phase 6 closure evidence pack (Wave 4)

Follow-up backlog (Phase 6 → Wave 5)

  • Per-step timeout_seconds override fully wired through executor
  • Real SkillDispatcher registry (replace trace-only stub)
  • Per-step env: allow-list with whitelist mapping
  • AC-1/AC-2/AC-10 real brownfield-code E2E + kill -9 resumability
  • /audit Round 1+2 на Phase 6 (security focus subprocess execution surface)
  • Activate PRD-072 / RFC-007 / ADR-010 (post-audit, R_eff ≥ 0.7 with EVID-090 + EVID-091)
  • Tag v0.27.0 + cargo-dist Release workflow
  • PR feat/phase6-real-dispatchers → dev (after audit)

Previous: v0.26.0 — PRD-065/066/067 Playbook + Ingest + Plugin detection (Phase 5)

EPIC-007 Phase 2 — Forgeplan становится оркестратором. ADR-009 implementation-complete: playbook runtime + ingest engine + plugin detection + canonical marketplace mapping/playbook. 4-wave sprint, 9 agents, ~9000 LOC.

PRD-065 — Playbook runtime + YAML schema

  • FR-1 Rust module forgeplan-core::playbook::{types,loader,executor,dispatch,journal}
  • FR-2 JSON Schema docs/schemas/playbook.schema.yaml
  • FR-3 CLI forgeplan playbook {list|show|run|validate}
  • FR-4 5-variant typed Delegation enum
  • FR-5 Step output capture via produces_at + mapping
  • FR-6 Journal .forgeplan/journal/playbook-runs.jsonl
  • FR-7 Progress reporting (TTY-aware stderr)
  • FR-8 Hint contract integration (PRD-071)
  • AC-1..AC-6 acceptance criteria covered (Wave 4 E2E)
  • Real Plugin/Agent/Skill subprocess dispatchers (closed by Phase 6 / PRD-072 FR-1..FR-5)

PRD-066 — Ingest engine + mapping YAML

  • FR-1 Rust module forgeplan-core::ingest::{types,sources,template,engine,idempotency}
  • FR-2 JSON Schema docs/schemas/mapping.schema.yaml
  • FR-3 CLI forgeplan ingest --mapping --source --dry-run
  • FR-4 Source-ref format {path}:{line_start}-{line_end}
  • FR-5 Idempotency via source_hash
  • FR-6 Canonical marketplace/mappings/c4-to-forge.yaml (Wave 4)
  • FR-7 forgeplan doctor --sources invariant validation
  • AC-1..AC-6 acceptance criteria covered (Wave 4 E2E)
  • MCP forgeplan_ingest wrapper (deferred — CLI cover via forgeplan serve)
  • 4 additional canonical mappings (autoresearch/git/ddd/spec → forge) — follow-up

PRD-067 — Plugin detection + self-describing hints

  • FR-1 Rust module forgeplan-core::plugins::{detection,registry,hints}
  • FR-2 Detection scanner paths (.claude/plugins/cache/, .agentskills/, etc.)
  • FR-3 Plugin registry с known plugins
  • FR-4 Project signal detector
  • FR-5 Playbook recommendation engine (signals × plugins → applicable)
  • FR-6 CLI forgeplan plugins {list|doctor|info}
  • FR-7 Hint extension (ADR-008 pattern + recommended_playbook)
  • AC-1, AC-2, AC-6 закрыты в Phase 5
  • AC-3 — forgeplan init empty repo → recommended: greenfield-kickoff (closed by Phase 6 / PRD-072 FR-6)
  • AC-4 — forgeplan init .obsidian/ → recommended: brownfield-docs (closed by Phase 6)
  • AC-5 — forgeplan init legacy code >100 commits → recommended: brownfield-code (closed by Phase 6)
  • AC-7 — backward compat (FORGEPLAN_HINTS=0 / non-TTY) (closed by Phase 6)

Documentation + Marketplace

  • docs/operations/PLAYBOOK-AUTHORING.ru.md (Wave 4 W4B)
  • docs/operations/INGEST-MAPPINGS.ru.md (Wave 4 W4B)
  • marketplace/mappings/c4-to-forge.yaml canonical (Wave 4 W4B)
  • marketplace/playbooks/brownfield-code.yaml canonical (Wave 4 W4B)
  • docs/README.md + docs/README.ru.md index entries (Wave 4 W4B)
  • CHANGELOG.md v0.26.0 section
  • EVID-089 — Phase 5 evidence pack

Follow-up backlog (Phase 5 — partially closed by Phase 6)

  • Real subprocess dispatch for delegate_to: plugin/agent/skill (closed by PRD-072 FR-1..FR-3 / Phase 6 Wave 1)
  • Canonical playbook greenfield-kickoff.yaml (closed by PRD-072 FR-7 / Phase 6 Wave 3)
  • MCP forgeplan_ingest wrapper (still deferred — CLI cover via forgeplan serve)
  • Canonical mappings: autoresearch / git-log / ddd / sparc → forge (still backlog)
  • Canonical playbooks: brownfield-docs.yaml, audit.yaml, release.yaml (3 of 4 still backlog)
  • Parallel step execution (DAG planner) — Non-Goals v1
  • Tag v0.26.0 + cargo-dist Release workflow (released in branch — pending tag)

Previous: v0.25.0 — PRD-071 hint contract (PR #212, 2026-04-27)

PRD-071 unified 5-rule hint contract shipped. Audit coverage 0% → 100% (70/70 CLI). 36 integration tests + drift-prevention audit script. Awaiting PR #212 review.

  • PRD-071 5-cycle multi-agent sprint complete (9 agents, 90 files)
  • EVID-086 linked, PRD-071 + PROB-046 active
  • CHANGELOG.md + docs/README + SKILL.md + agent-protocol.md updated
  • PR #212 merged to dev (awaiting review)
  • Tag v0.25.0 after release PR

Deferred from PRD-071 (low priority)

  • FR-013: forgeplan health add "Hint coverage" metric (audit script sufficient for now)
  • FR-011: separate forgeplan-mcp/tests/hint_contract.rs integration test file (covered via unit tests + dogfood)
  • FR-014: RU localization for hints (backlog)

Follow-up items

  • PROB-047 (potential): scan-import should dedupe by title/content hash — current behavior creates duplicate PRDs from doc files

Previous: v0.24.0 released 2026-04-19 — PRD-057 multi-agent dispatcher

Next priorities (from ROADMAP)

  • Sprint A: Public Presence — Website (PRD-024) + README + crates.io + Docker
  • Sprint B: CI/CD Integration — validate/health --ci + GH Action
  • Sprint C: Desktop App — EPIC-004 (Tauri + React)
  • Sprint D: Ecosystem — VS Code ext + GitHub Issues bridge

Open bugs

  • PROB-026: tag canonicalization (PR #169 pending merge)
  • PROB-027: reindex without lance/ (PR #169 pending merge)
  • PROB-035 remainder: code-fence awareness in extract_field

Previous: v0.17.2 quality hotfix 2026-04-09 — E2E verification sprint

v0.17.2 hotfix P0

  • PROB-030 BM25 prefix fallback (smart.rs max(bm25_norm, kw))
  • PROB-031 score.rs imports core parser (deleted local duplicate with CL0 default)
  • PROB-032 breakdown display (auto-fixed by PROB-030)
  • PROB-033 new evidence phase-agnostic (no session warning)
  • PROB-034 CRITICAL multi-line HTML comment state machine in extract_field
  • F1 fail-closed on unclosed <!-- (unclosed → CL0 + warn)
  • F2 fail-closed on unparseable congruence_level (garbage → CL0 + warn)
  • evidence template simplified (single-line comments only, no booby-trap)
  • 12 new regression tests (total 1143 pass, +12 from 1131)
  • Cargo.toml workspace + cross-crate refs → 0.17.2
  • CHANGELOG.md v0.17.2 entry
  • CLAUDE.md status block updated to v0.17.2
  • TODO.md updated (this block)
  • 4-agent audit completed (A code, B tests, C security, D docs)
  • All audit blockers addressed in-scope
  • PROB-034 card + EVID-068..072 created
  • EVID-068..072 Interpretation + CL Justification filled (audit D)
  • PROB-030..034 + EVID-068..072 activated
  • PR release/v0.17.2 → main (#163 merged)
  • Tag v0.17.2 + push (cargo-dist Release workflow success, brew formula published)
  • Sync main → dev via PR (#164 merged)
  • Health debt cleaned: 8 stubs deprecated/superseded, 5 dup EVIDs deprecated, 3 orphan NOTEs linked
  • Hindsight retain v0.17.2 finale
  • PROB-035 "extract_field hardening" filed for follow-up sprint (code-fence, token-boundary substring)

v0.17.1 hotfix ✅ (shipped)

  • PROB-028 phantom rows (PRD-044)
  • PROB-029 health verdict (PRD-045)
  • CHANGELOG + tag + cargo-dist + sync done

Previous: v0.17.0-rc — EPIC-003 complete, ready to tag

Stats (v0.17.0)

  • ~56 CLI commands, ~47 MCP tools, 1109 tests (+280 from v0.16)
  • Workspace: 0 warnings on both default and --features semantic-search
  • ~13.8K LOC added across EPIC-003 (Sprints 13.0 → 13.7 + post-closeout hotfix)
  • PRs #141-#156
  • E2E: sprint-13.6-regression.sh (16 checks) + sprint-13.7-regression.sh (16 + SEMANTIC_E2E opt-in), 0 failures
  • LLM: gemini-3-flash-preview
  • Distribution: cargo-dist v0.31.0, 5 targets, brew + install.sh + checksums
  • Pipeline: Shape→Validate→ADI→Code→Test→Fmt→Lint→Audit→Fix→Re-audit→Manual UX→Closeout
  • Sprint Checklist Framework (NOTE-044) landed 2026-04-08 as reusable quality gate
  • ADI mandatory for Standard+ depth (CLAUDE.md methodology update)

v0.17.0 done — EPIC-003 Search, Discovery, Intelligence ✅

  • Sprint 13.0 Security + ADR-007 (2h, no artifact)
  • Sprint 13.1 PRD-043 Methodology Integrity — EVID-058, PR #145
  • Sprint 13.1.5/.7 Hardening + integrity config wiring — EVID-059
  • Sprint 13.2 PRD-039 Smart Search v2 (BM25 + filter DSL + graph expansion) — EVID-065 (backfill during final audit)
  • Sprint 13.3 PRD-035 p1 Tags + Source Tier — EVID-060
  • Sprint 13.4 PRD-035 p2 Discover MCP tools + CLI — EVID-061
  • Sprint 13.5 PRD-040 Scoring Intelligence (Skills Memory + R_eff CI) — EVID-062, PR #153
  • Sprint 13.6 PRD-041 FPF Rules CLI + MCP — EVID-063, PR #154
  • Sprint 13.7 PRD-042 FPF KB Vector Search (supersedes PRD-018) — EVID-064, PR #155
  • Sprint 13.7 post-closeout hotfix — 19 debts triaged, 11 fixed, NOTE-044/045 (PR #156)
  • Final release audit — 4 parallel auditors, version bump, CHANGELOG, bugfix agent
  • PRD-018 superseded by PRD-042, EPIC-003 activated

P0: Release v0.17.0 tag

  • Cargo.toml version bump 0.16.0 → 0.17.0 (workspace + 3 path deps)
  • CHANGELOG.md created with full v0.17.0 entry
  • 7 EVIDs activated (058..064) + EVID-065 backfill for Sprint 13.2
  • PRD-039 activated (R_eff=1.00, F-G-R=0.88 A)
  • EPIC-003 activated
  • Title validation bugfix (fa97f10, tag-prep-bugfix agent)
  • Commit tag-prep changes (6a1904f)

P1: After release PR merged

  • PR release/v0.17.0 → main (merge commit)
  • Tag v0.17.0 + push
  • Sync main → dev
  • Hindsight memory_retain EPIC-003 finale

P0: FPF Engine v2 Phase 2 — Sprint 12 (RFC-001) ✅

  • ADI reasoning: H2 Two-tier Rules selected (FPF B.5.2 Abductive Loop)
  • ext/rules.rs: Rule engine with expressions, graph-aware, time-aware (~600 LOC)
  • Dashboard integration: rule engine replaces explore::suggest, HashMap O(N+R)
  • Bounded context in reason output (CLI + MCP)
  • Config template with rule examples in forgeplan init
  • FpfConfig.rules field (empty = default 5 rules)
  • 4 audit agents: code review, bounded context, Rust expert, security
  • Audit fixes: NaN rejection, empty condition guard, circular scoring, TOCTOU, O(N+R)
  • 38 rule engine tests (unit + scenario + negative + corner), 829 total
  • EVID-057 linked, R_eff=1.00, PRs #133 + #135 merged
  • KB vector search — deferred to Phase 3+ (keyword works, NOTE-039 DSL idea)

P0: FPF Engine v2 Phase 1 — Sprint 11 (RFC-001) ✅

  • EPIC-002 shaped + activated (PR #128)
  • RFC-001 shaped: 3 options, ADI confirmed Option C (Layered Core+Ext)
  • fpf/core/ module: config.rs, trust.rs, adi.rs, model.rs (34 tests)
  • FpfConfig wired into CLI scoring (score, fgr, context, dashboard)
  • Config templates in init + current config.yaml (all 6 sections)
  • Audit: 3 agents, 3H + 1M fixed, NaN validation, reliability clamp
  • EVID-055, R_eff=1.00, RFC-001 activated
  • Housekeeping: 12 orphans linked, SPRINTS.md updated
  • 1.5: AdiRecord wiring — reason --save creates structured JSON in Note body
  • 11.3: CI/CD Linter — health --ci + validate --ci + GH Actions workflow (PR #132)
  • PR #131 merged + progress synced

P0: Distribution Pipeline — Sprint 10 (PRD-023) ✅

  • PRD-023 shaped + validated (8 FR, 4 journeys, 4 AC, Deep depth)
  • ADI reasoning: 3 hypotheses → H1 cargo-dist selected (High confidence)
  • cargo-dist v0.31.0 integrated (dist-workspace.toml, release.yml generated)
  • 5 targets: macOS arm64/x86, Linux x86/musl, Windows x86
  • Installers: shell (install.sh) + Homebrew (AiDogfood/homebrew-tap)
  • Cargo.toml metadata: homepage, keywords, categories for crates.io
  • 2-agent audit: 4C + 3H + 3M findings → all fixed
  • Action versions @v6/@v7 → @v4 (cargo-dist v0.31.0 bug)
  • .gitignore: dist manifest files added
  • Embed fix resolved: fastembed v5.13.0 compiles (upstream fixed)
  • EVID-050 active, R_eff=0.80, PR #97 merged
  • 753 tests, 0 failures, project healthy

P0: ADI Quality + LLM + E2E + Cleanup — Sprint 9 (PROB-021) ✅

  • PROB-021: ADI prompt enriched (metadata, relations+titles, architecture hint)
  • System prompt: justified confidence, project context awareness
  • reason_temperature config field, architecture hint from file
  • evidence_needed → CLI "Next steps" UX
  • Model benchmark: 4 models × 7 artifacts → gemini-3-flash-preview selected
  • E2E Wave 8 (LLM): 10/10 pass (generate, reason, decompose, capture, context)
  • Draft cleanup: 34→12 (7 deleted, 15 deprecated)
  • cargo fmt entire codebase (122 files) + pre-commit-fmt.sh hook
  • Pipeline updated: +Fmt step, ADI mandatory for Standard+
  • 6 new E2E integration tests (cascade delete, lifecycle, deprecated blocking)
  • EVID-048 active, R_eff=0.90, PR #96 merged
  • 753 tests, 0 failures, project healthy

P0: Graph Integrity — Sprint 8 (PROB-020) ✅

  • BUG-1 (P1): blocked/order treated deprecated as blockers → resolved_ids
  • BUG-2 (P1): delete cascade relations + phantom PROB-013 cleanup
  • BUG-2b: unlink resilient for phantom relations
  • 5-agent audit: 2 critical + 5 warnings → all fixed
  • 2 new MCP tools: forgeplan_blocked + forgeplan_order
  • validate_id_for_filter() whitelist, DRY common::resolved_ids()
  • O(n²)→O(n) order.rs, double scan eliminated, TOCTOU fixed
  • route "" rejects empty, memory excluded from orphan detection
  • 83 E2E commands tested, E2E-TEST-PLAN.md created
  • PROB-020 active, EVID-047 linked, R_eff=1.00, PR #95

P0: E2E Bug Fixes — Sprint 2 (PROB-018)

  • BUG-001 (P1 Security): scan --path /tmp path traversal — added project root boundary validation (coverage.rs)
  • BUG-002 (P2): unlink не проверяет существование связи — added existence check (link.rs)
  • BUG-003 (P3): lifecycle transition message "draft → active" hardcoded — uses old_status (activate.rs)
  • 2 new unit tests for delete_relation (store.rs)
  • Create Evidence EVID-040 + link to PROB-018
  • Audit (4-agent: logic+security+rust+task) + PR #85 merged

P0: Lifecycle v2 — Sprint 3 (ADR-005) ✅

  • ADR-005 shaped + validated (new state machine design)
  • Phase 1: stale status + terminal deprecated/superseded (transitions.rs, 15 tests)
  • Phase 2: renew() + reopen() core logic (lifecycle/mod.rs, 10 tests)
  • Phase 3: CLI commands (renew.rs, reopen.rs)
  • PROB-019: self-link guard (store.rs + in_memory.rs, 4 tests)
  • 5-agent audit: 8 findings fixed (date validation, reason sanitization, atomicity)
  • EVID-041 (PROB-019) + EVID-042 (ADR-005) linked + activated
  • ADR-005 activated, R_eff = 0.80
  • PR #85 merged

P0: Estimate Engine (PRD-022) ✅

  • PRD-022 shaped + validated (8 FR, 3 journeys)
  • RFC-005 architecture (3 phases, 12 tasks)
  • ADR-004 hybrid approach (rule-based L0 + LLM L1)
  • Phase 1: types, extractor, scorer, calculator, display (35 tests)
  • Phase 2: confidence, CLI estimate command (39 tests)
  • 2-agent audit: 2 CRITICAL + 4 HIGH fixed (50 tests)
  • Config integration: grade_profile, multipliers, --my-grade
  • FORGEPLAN-GUIDE: Estimate Engine section
  • Evidence EVID-036 linked, PRD-022 + RFC-005 + ADR-004 activated

P0: MemoryDriver (RFC-003 Phase 2) ✅

  • remember/recall CLI commands
  • ArtifactKind::Memory with mem- prefix
  • DRY helpers in common.rs + 3 tests

P0: Quality Cycle Sprint (v0.12.0) ✅

  • 17 MUST gaps → 0 (enable --depth update)
  • Evidence parser: ignores template placeholders (CL0→CL3)
  • LLM scorer (RFC-005 Phase 3): --llm-score flag, PR #79
  • Hints system: 9 commands, 11 tests, shared hints.rs
  • Domain inference: frontmatter → keywords → LLM (3-level)
  • Manual complexity override: --complexity "FR-001=8"
  • Spec/Evidence confidence boost: +15%/+20%
  • forgeplan link body reset fix (PR #75)
  • dotenvy for .env API keys
  • Template FR filtering, keyword TaskType improvements
  • 40 commands smoke-tested
  • Release v0.12.0 tagged + installed

P0: Integrity Issues (PROB-012 dogfood audit) ✅

  • Semantic search broken — feature flag propagated CLI→core via Cargo.toml
  • R_eff divergence — update_r_eff_score() persists to LanceDB, NaN guard
  • health vs journal inconsistency — expanded blind spots + decision kinds aligned
  • coverage 0% — Affected Files section added to templates + backfilled 18 active PRD/RFC/ADR
  • route underestimates — 4 keyword triggers + 2 heuristics added

Fixed in commit d84bc69 (fix/prob-012-integrity-remediation). 2 audit rounds, 403 tests.

Known Issues

  • changelog commit_hash: LanceDB schema migration — old workspaces lack commit_hash column. forgeplan update logs warning. Fix: reinit workspace or add column migration.
  • RFC-005 3.2: estimate MCP tool — grade param wired (Sprint 1 housekeeping).
  • 1 STUB artifact: unidentified, low priority.
  • e2e_coverage_backfill test: pre-existing failure, unrelated to v0.12 changes.
  • Self-link guard (PROB-019): link X X rejected with "Self-link not allowed" (Sprint 3).
  • Runbook outdated (NOTE-031): deprecated — file doesn't exist in repo, discrepancies noted in TODO.
  • LLM tests: Wave 8 (10 commands) passed with gemini-3-flash-preview. Wave 10 edge cases still pending.
  • --semantic feature flag: search --semantic fails at runtime if not compiled with semantic-search.

Open Problems

ID Priority Title Status
PROB-001 Done Data loss → solved by PRD-009 export/import ✅
PROB-002 P2 Auth reuse — separate API key barrier Open
PROB-003 Done Dead statuses → solved by PRD-007 lifecycle ✅
PROB-004 Done Agent drift → solved by PRD-010 hooks ✅
PROB-005 Done Cold start → solved by PRD-012 init --scan ✅
PROB-006 Deprecated Routing misses UX scope → fixed v0.11, keywords expanded ✅
PROB-009 Deprecated F-G-R Granularity → future PRD scope ⚠️
PROB-010 Deprecated Markdown projections → fixed by RFC-004 files-first ✅
PROB-012 Deprecated Feature integrity gap → 5 fixes, 2 audit rounds ✅
PROB-013 Deleted R_eff skip non-active → implemented in ADR-002, deleted ✅
PROB-014 Deprecated Smart search gaps → fixed v0.12, real cosine ✅
PROB-016 Deprecated CLI quality → 13 fixes, 6-agent audit ✅
PROB-018 Done E2E Smoke Test Findings — 3 bugs fixed, 4-agent audit, PR #85 ✅
PROB-019 Deprecated Self-link guard added — case-insensitive check, 4 tests ✅
PROB-020 Done Graph integrity — 10 bugs, 5-agent audit, cascade delete, PR #95 ✅
PROB-021 Done ADI quality — enriched prompt, model benchmark, fmt hooks, PR #96 ✅

Backlog (приоритизированный)

P1: Release v0.13.0 (Distribution)

  • Tag v0.13.0 → trigger first automated release via GH Actions
  • Verify brew install + install.sh on clean machine
  • cargo publish (manual, safety hook blocks)

P1: Embed & Distribution ✅

  • Embed feature fix — fastembed v5.13.0 compiles, upstream resolved
  • Distribution — cargo-dist v0.31.0, PR #97 merged (Sprint 10)

P2: Integrity Follow-up (from FPF audit) ✅

  • Read-back verify in update_r_eff_score — pre-check with get_record before update
  • DRY decision_kinds — DECISION_KINDS_EVIDENCE + DECISION_KINDS_JOURNAL in types.rs
  • Coverage batch-update — forgeplan coverage --backfill (18 artifacts updated)
  • PROB-013 — R_eff skip deprecated/draft in recursive chain (ADR-002)
  • Tree visual — evidence/note show ·· instead of 0.00
  • METHODOLOGY-COURSE.md — Chapter 8 added (tree, coverage, hooks, R_eff rules)
  • PRD-019 Layer 3 — MCP session state machine (backlog, PRD-019 activated)

P2: Route & Enforcement (from usability testing)

  • Route gap: added "new command/feature" keywords (English)
  • Batch score CLI: forgeplan score --all implemented
  • LLM-first route: 3-level routing (L0 keywords, L1 LLM classify, L2 FPF ADI reasoning) — PRD-020, 444 tests
  • PRD-019 Layer 3: MCP session state machine — агент не может пропустить Shape phase
  • Duplicate notes cleanup: NOTE-004 deprecated (duplicate of NOTE-005)

P1: Smart Search (PROB-014, v0.12)

  • F1 (P0): Embed title + body snippet — embedding_text() + forgeplan embed command
  • F2 (P0): Graph walk shows relation types — neighbors_with_relations()
  • F3 (P1): forgeplan embed — batch embed all artifacts (persistent in LanceDB)
  • F4 (P1): Smart search — text-first + boosters (Algolia-style, not weighted sum). EVID-033.
  • F5 (P1): forgeplan gaps — 18 MUST gaps found on real data! Audit: 4 fixes.
  • F6 (P2): Fix evidence blind spots (EVID-015, EVID-025, EVID-026, EVID-027)
  • forgeplan search --semantic — vector-only search
  • forgeplan search — smart by default (keyword + semantic + R_eff + status + graph boosters)
  • forgeplan search --keyword — forced keyword grep
  • Configurable embedding model via config.yaml (BGE-M3 default)
  • Configurable chunk_size via config.yaml (default 2000)
  • Future: Reciprocal Rank Fusion (RRF) for production-grade hybrid search

P0: CLI Quality Remediation (PROB-016, 3-agent deep audit) ✅

Wave 1 — BROKEN (6-agent team sprint, PR #65):

  • B1: deprecate --reason stores reason in body (## Deprecation section)
  • B2: update --status active blocked — must use forgeplan activate
  • B3: 4 LLM commands — pre-flight API key check via require_llm_config()
  • B4: review checks evidence+stub gates (same as activate)

Wave 2 — SAFETY:

  • N1: delete checks dependents, warns + requires --yes
  • N7: supports added to VALID_RELATIONS
  • N8: init --force warns about data loss
  • N9: unlink updates projection

Wave 3 — CORRECTNESS:

  • N2: new depth=tactical for note/evidence/problem/solution/refresh
  • N3: supersede warns if replacement already superseded/deprecated
  • N4: score --all --json clean JSON output
  • N5: update --depth bails with error
  • N6: order/blocked structural relations only (informs doesn't block)

EVID-034. 532 tests. Deferred: fgr/blindspots redundancy, graph filtering, drift adoption, export embeddings

P1: Driver Abstraction — RFC-003

  • Phase 1: StorageDriver trait + LanceDriver + InMemoryStore + factory — PR #61 merged
  • Phase 1 audit: 3 agents, 13 findings, 7 fixed (C1-C3, H1, H3, M1, M2)
  • Phase 1 deferred (PROB-015): H2 EmbedDriver, H4 ISP split, M3-M5, test gaps
  • Phase 2: MemoryDriver (remember/recall) — PR #72 merged
  • Phase 3: SQLite driver + feature flags
  • Phase 4: Config-driven selection + forgeplan init shows drivers

P2: Architecture — Files as Source of Truth (ADR-003, RFC-004) ✅

  • Invert direction: .md files = truth, LanceDB = index (RFC-004 Phase 1, PR #67)
  • File watcher (notify crate) for auto-reindex (RFC-004 Phase 2, PR #69)
  • forgeplan reindex — one-time full re-sync from .md files (PR #71)
  • R_eff computed on-the-fly from evidence files (not stored)
  • Links in frontmatter related: field (RFC-004 Phase 1)
  • Change log tracking (RFC-004 Phase 3, PR #69)
  • Git-sync integration (RFC-004 Phase 4, PRs #80-#81)

P2: Polish

  • Binary size optimization — release profile 163MB→41MB (-75%)
  • fpf.rs миграция на common::store() (6 functions)
  • coverage.rs, scan_import.rs миграция на common::open_store()

P2: CLI UX Polish (NOTE-029, from E2E findings)

  • forgeplan links PRD-001 — show all relations for an artifact (1 day)
  • forgeplan validate --ci — exit 1 on MUST errors for CI/CD (1 day)
  • forgeplan doctor — check workspace, LLM key, feature flags (2 days)
  • Document capture as LLM-dependent in --help
  • Error consistency: choose idempotent vs strict philosophy
  • Document case-sensitive IDs

P2: Agent Memory Engine (NOTE-025, Direction A — HIGH R_eff)

  • Test forgeplan serve as MCP server in Claude Code
  • Claude Code plugin: /fp-validate, /fp-context, /fp-score skills
  • capture offline mode (create Note/ADR without LLM)
  • forgeplan watch --emit-events — JSON event stream for agents

P2: CI/CD Architecture Linter (NOTE-026, Direction B)

  • forgeplan health --fail-on — configurable thresholds
  • GitHub Action: uses: forgeplan/action@v1

P3: Ruflo/Gastown Integration (NOTE-027)

  • MCP config example for Ruflo (.agents/config.toml)
  • Architecture-guardian custom agent YAML
  • Gastown directive template

P3: Task Tracker Bridges (NOTE-028)

  • Bidirectional sync with task trackers (Linear, Jira, Orchestra)
  • Export to GitHub Issues / Linear tasks
  • Webhook on activate/supersede events

Phase 5: Desktop App

  • Tauri 2.0 + React frontend (shared Rust core)

Done ✅

  • v0.10.1 — Hotfix: bidirectional R_eff evidence lookup
  • v0.10.0 — scan-import, --json x14, CLI UX, audit fixes, security hardening
  • v0.9.0 — PRD-016..021: R_eff recursive, BMAD v2, OpenSpec DAG, FPF KB, codebase awareness, decision contracts
  • v0.8.0 — CLI UX: cliclack init, styled output, setup-skill
  • v0.7.0 — EPIC-001 complete, FPF engine, lifecycle, /forge skill
  • v0.6.0 — Methodology Engine: routing, lifecycle, F-G-R
  • v0.5.0 — Health, Journal, Validation v2
  • Phase 4 — MCP Server + AI Features + CRUD
  • Phase 3 — Core CLI + LanceDB Primary
  • Phase 1 — Schemas, Templates & Docs
  • Phase 0 — Foundation & Research

v0.10.x detailed

  • PRD-012: Project Onboarding — init --scan, scan-import, 3-tier detection
  • PRD-008: CLI UX — 8 ui helpers, --json for 14 commands, common::store() (-128 LOC)
  • 4-agent Rust audit: UTF-8 safety, symlink protection, file size limits, import validation
  • R_eff bidirectional evidence fix (get_incoming_relations)
  • All blind spots (3), orphans (8), at risk (11) → resolved
  • 14 draft evidence → activated