diff --git a/.planning/formal/model-registry.json b/.planning/formal/model-registry.json index bc44e01db8..27c95b2b6e 100644 --- a/.planning/formal/model-registry.json +++ b/.planning/formal/model-registry.json @@ -1,6 +1,6 @@ { "version": "1.0", - "last_sync": "2026-03-27T08:57:20.140Z", + "last_sync": "2026-03-27T09:56:34.577Z", "models": { ".planning/formal/alloy/account-pool-structure.als": { "version": 1, @@ -1335,8 +1335,8 @@ "consecutive_pass_count": 670 }, ".planning/formal/alloy/quorum-votes.als": { - "version": 2359, - "last_updated": "2026-03-27T08:57:19.964Z", + "version": 2361, + "last_updated": "2026-03-27T09:56:34.440Z", "update_source": "generate", "source_id": "generate:formal-specs", "session_id": null, @@ -2308,8 +2308,8 @@ "consecutive_pass_count": 670 }, ".planning/formal/prism/quorum.pm": { - "version": 2359, - "last_updated": "2026-03-27T08:57:19.966Z", + "version": 2361, + "last_updated": "2026-03-27T09:56:34.441Z", "update_source": "generate", "source_id": "generate:formal-specs", "session_id": null, @@ -2571,16 +2571,16 @@ "consecutive_pass_count": 670 }, ".planning/formal/tla/NFQuorum.tla": { - "version": 2359, - "last_updated": "2026-03-27T08:57:19.962Z", + "version": 2361, + "last_updated": "2026-03-27T09:56:34.438Z", "update_source": "generate", "source_id": "generate:formal-specs", "session_id": null, "description": "" }, ".planning/formal/tla/NFQuorum_xstate.tla": { - "version": 1632, - "last_updated": "2026-03-27T08:57:20.140Z", + "version": 1634, + "last_updated": "2026-03-27T09:56:34.577Z", "update_source": "generate", "source_id": "generate:formal-specs", "session_id": null, diff --git a/.secrets.baseline b/.secrets.baseline index cb819ae413..2321f8a1fe 100644 --- a/.secrets.baseline +++ b/.secrets.baseline @@ -240,7 +240,7 @@ "filename": "bin/manage-agents-core.cjs", "hashed_secret": "32df82dccdd26b62d92eaefa9ccc2550d5ec96bd", "is_verified": false, - "line_number": 393 + "line_number": 406 } ], "bin/manage-agents.test.cjs": [ @@ -391,5 +391,5 @@ } ] }, - "generated_at": "2026-03-26T13:25:00Z" + "generated_at": "2026-03-27T10:10:42Z" } diff --git a/CHANGELOG.md b/CHANGELOG.md index d6371f5768..2a051fcc2b 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -6,63 +6,141 @@ Format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/). ## [Unreleased] +## [0.41.3] - 2026-03-27 — Live Health Dashboard Fix + +Incremental on [0.41.2]. Fixes TUI health dashboard that silently skipped all subprocess providers. + +### Added +- `feat(agents)`: subprocess probing and `providers.json`-backed dashboard — TUI Live Health dashboard now probes subprocess providers via CLI `--version` checks and shows provider name + model from `providers.json` + +### Fixed +- `fix(tui)`: remove duplicate `pdata` declaration in `checkHealthSingle` — eliminated early bail that blocked probing subprocess providers entirely + ## [0.41.2] - 2026-03-27 — Enhanced Resolve Triage +Incremental on [0.41.1]. Overhauls `/nf:resolve` presentation with structured analysis and quorum integration. + ### Added - **Detailed resolve reports** — `/nf:resolve` now shows Key Files, Analysis, Pros & Cons, and Conclusion sections for every item type (solve items, pairings, orphan models, orphan requirements) -- **Clickable file paths** — All file references in resolve output use `file_path:line_number` format for direct navigation -- **Mandatory quorum action bar** — Every resolve item and batch displays `[q] Quorum` option for multi-model consensus review -- **Batch action choices** — Batch presentations include confirm-all, FP-all, and individual-item-by-number actions +- **Clickable file paths** — all file references in resolve output use `file_path:line_number` format for direct navigation +- **Mandatory quorum action bar** — every resolve item and batch displays `[q] Quorum` option for multi-model consensus review +- **Batch action choices** — batch presentations include confirm-all, FP-all, and individual-item-by-number actions ### Changed -- **Resolve presentation format** — Upgraded from simple verdict+recommendation to structured analysis with pros/cons trade-offs per action option -- **Orphan model context** — Shows 15-20 lines of model file (up from 10-15) with module identification -- **Pairing analysis depth** — Reads both model file and requirement text to assess semantic connection vs keyword overlap +- **Resolve presentation format** — upgraded from simple verdict+recommendation to structured analysis with pros/cons trade-offs per action option +- **Orphan model context** — shows 15-20 lines of model file (up from 10-15) with module identification +- **Pairing analysis depth** — reads both model file and requirement text to assess semantic connection vs keyword overlap ### Fixed -- **Local patch drift** — Synced enhanced resolve.md from nf-local-patches back to repo source +- **Local patch drift** — synced enhanced resolve.md from nf-local-patches back to repo source +- `fix(ci)`: skip state-space guard for MCMCPEnv bounded model — individual step + master FV runner + +## [0.41.1] - 2026-03-26 — Risk-Based Adaptive Quorum, Solve Reporting & Diagnostic Sweeps -## [0.41.1] - 2026-03-26 — Risk-Based Adaptive Quorum Fan-Out +Incremental on [0.41.0]. Adds risk-aware quorum sizing, expands diagnostic sweeps to 20 layers, and adds automation-first verification. Includes quick tasks 354-361. ### Added -- **Risk classifier** — Haiku subagent in quick workflow Step 2.7 classifies tasks as low/medium/high risk based on file count, task type, requirements impact, and scope + +#### Risk-Based Adaptive Quorum (quick-360) +- `feat(quick-360)`: add risk-based adaptive quorum fan-out with Haiku risk classifier — classifies tasks as low/medium/high risk based on file count, task type, requirements impact, and scope - **Adaptive quorum fan-out** — Step 5.7 dispatches low=1 (skip quorum), medium=3, high=5 participants based on risk level -- **`--force-quorum` flag** — Overrides low risk classification to medium, forcing external quorum dispatch -- **Quorum audit logging** — Structured audit log emitted for every quorum reduction or skip with risk level, reason, and fan-out count +- **`--force-quorum` flag** — overrides low risk classification to medium, forcing external quorum dispatch +- **Quorum audit logging** — structured audit log emitted for every quorum reduction or skip with risk level, reason, and fan-out count - **Risk guardrails** — ROADMAP, formal model, hook, and workflow files can never be classified as low risk - **Scope contract risk fields** — `risk_level` and `risk_reason` persisted in scope-contract.json -- **Graph-first discovery** — formal-scope-scan.cjs uses unified semantic+graph search module shared with candidate-discovery.cjs -- **Automation-first verification** — verify-work and execute-phase workflows prefer Playwright/agent-browser over manual testing + +#### Solve Pipeline Enhancements (quick-354 through quick-357) +- `feat(quick-354)`: add 5 missing layers to solve-report table renderer — full 20-layer coverage with checker alignment and signals fixes +- `feat(quick-355)`: auto-invoke `/nf:resolve` after solve finishes iterating +- `feat(quick-356)`: add 7 new sweep functions and fold 8 scripts into existing sweeps — wire 7 new sweeps into `computeResidual`, totals, table, `DEFAULT_WAVES` +- `feat(quick-357)`: add `@requirement` annotations to 8 domain-named test files + require-path tracing to `sweepTtoR` (TC-CODE-TRACE-8 test) + +#### Formal & Discovery (quick-358, quick-359, quick-361) +- `feat(quick-358)`: extract unified graph search module and add graph-first discovery to `formal-scope-scan.cjs` — shared with `candidate-discovery.cjs` +- `feat(quick-359)`: allow `formal_artifacts: create` when scope-scan returns empty — enables new model bootstrapping +- **Automation-first verification** (quick-361) — verify-work and execute-phase workflows prefer Playwright/agent-browser over manual testing ### Changed -- **quorum-dispatch.md Section 3** — Updated fan-out mapping: low=1/skip, medium=3, high=5 (previously low=2, high=MAX_QUORUM_SIZE) -- **Quorum timeout defaults** — Increased from 30s to 300s for slot worker dispatch +- **quorum-dispatch.md Section 3** — updated fan-out mapping: low=1/skip, medium=3, high=5 (previously low=2, high=MAX_QUORUM_SIZE) +- **Quorum timeout defaults** — increased from 30s to 300s for slot worker dispatch ### Fixed -- **Formal scope scan gap** — Allow `formal_artifacts: create` when scope-scan returns empty (new model bootstrapping) -- **Orphan requirements** — Triaged 48 orphan requirements across 6 category groups +- `fix(ci)`: allowlist quorum debate transcripts in gitleaks +- `fix(ci)`: fetch latest gitleaks version dynamically +- `fix(ci)`: use gitleaks-action v2 and unpin detect-secrets +- `fix(quick-354)`: revise plan to address checker alignment and signals issues +- `fix`: use portable nf-bin path in debug.md require() for lint-isolation +- **Orphan requirements** — triaged 48 orphan requirements across 6 category groups ## [0.41.0] - 2026-03-25 — Unified Autoresearch Execution Pipeline -### Added -- **Debug pipeline absorption** — `/nf:debug` absorbs model-driven-fix Phases 1-4 as Steps A.5-A.8 (discovery, reproduction, refinement via Loop 1, constraint extraction) -- **Constraint injection** — Formal constraints from Loop 1 injected as [FORMAL CONSTRAINTS] block into quorum worker prompts -- **Formal model artifact tracking** — Debug artifact includes reproducing model path, verdict, constraints, and TSV trace -- **Task classification** — Haiku subagent at quick.md Step 2.7 classifies tasks as bug_fix/feature/refactor with confidence scoring -- **Bug fix routing** — Step 5.8 routes bug_fix tasks (confidence >= 0.7) through `/nf:debug` before executor -- **Debug context injection** — Executor prompt receives conditional debug_context block with constraints and formal verdict -- **Loop 2 pre-commit gate** — simulateSolutionLoop with onTweakFix fires before commit in both quick.md and execute-phase.md executors -- **Fail-open/strict modes** — Loop 2 gate warns by default (fail-open), blocks with --strict flag +Full milestone release building on [0.40.2]. Includes 4 milestone phases (50-53) and 17 quick tasks (337-353). Major themes: autoresearch-style iteration across all formal loops, solve pipeline optimization, debug unification, and `/nf:model-driven-fix` deprecation. + +### Added — Milestone Phases + +#### Phase 50: Debug Integration +- `feat(50-01)`: rewrite debug.md Steps A.5-A.8 to absorb model-driven-fix Phases 1-4 +- `feat(50-02)`: wire constraint injection into quorum worker prompts and add formal model artifact tracking — constraints from Loop 1 injected as `[FORMAL CONSTRAINTS]` block + +#### Phase 51: Task Classification & Debug Routing +- `feat(51-01)`: add task classification subagent to quick.md Step 2.7 — Haiku classifies tasks as bug_fix/feature/refactor with confidence scoring +- `feat(51-02)`: add debug routing (Step 5.8) and debug context to executor prompt — routes bug_fix tasks (confidence >= 0.7) through `/nf:debug` before executor - **Classification persistence** — scope-contract.json extended with classification object (type, confidence, routed_through_debug) -### Changed -- **solve-remediate b_to_f** — Rewired from `/nf:model-driven-fix` to `/nf:debug` dispatch +#### Phase 52: Loop 2 Pre-Commit Simulation Gate +- `feat(phase-52)`: add Loop 2 pre-commit simulation gate to both executor workflows (GATE-01..04) — `simulateSolutionLoop` with `onTweakFix` fires before commit in quick.md and execute-phase.md +- **Fail-open/strict modes** — Loop 2 gate warns by default (fail-open), blocks with `--strict` flag + +#### Phase 53: Skill Deprecation +- `feat(phase-53)`: replace model-driven-fix with deprecation shim (DEPR-01) — directs users to `/nf:debug` +- `feat(phase-53)`: rewire solve-remediate b_to_f layer to `/nf:debug` (DEPR-02) + +### Added — Quick Tasks + +#### Solve Loop Optimization (quick-337 through quick-346) +- `feat(quick-337)`: fast-path initial diagnostic in solve orchestrator — skips redundant re-scan when no residual exists +- `fix(quick-338)`: exit 0 on successful diagnostic, add `has_residual` JSON field +- `feat(quick-339)`: create `solve-inline-dispatch.cjs` for pre-running trivial layers without full Agent spawn + wire into solve orchestrator and remediation +- `feat(quick-340)`: conditional Haiku classification with 4 skip conditions — avoids redundant classify calls in solve-classify +- `feat(quick-341)`: cascade budget for R→F remediation dispatch — prevents unbounded cascades +- `feat(quick-341)`: add anti-self-answer guard + question-file + nonce to quorum dispatch +- `feat(quick-342)`: N-layer cycle detection with state hashing and bounce counting — detects remediation oscillation across layers +- `feat(quick-343)`: parallelize F→C sweep via background pre-spawn for faster diagnostics +- `feat(quick-344)`: incremental diagnostics by file delta — only re-scans files changed since last diagnostic run +- `feat(quick-345)`: two-phase solve with `--plan-only` and `--execute` flags +- `feat(quick-346)`: persistent solve state with `--resume` and iteration logging across context resets + +#### Formal Model Enhancements (quick-347 through quick-353) +- `feat(quick-347)`: add `formal-coverage-intersect.cjs` with tests + `--sync` mode and executor auto-detection wiring +- `feat(quick-348)`: add `autoresearch-refine.cjs` with tests + wire into model-driven-fix and solve-remediate +- `feat(quick-350)`: add `onTweakFix`, rollback, TSV trace, when-stuck behaviors to solution-simulation-loop + update model-driven-fix Phase 4.5 from CLI to require() +- `fix(351)`: enforce FALLBACK-01 in all workflow fail-open rules and add preflight slot/fallback preview display +- `feat(quick-352)`: add TLC process timeout and model size guards to formal verification spawning +- `feat(quick-353)`: add state-space preflight guard to `run-tlc.cjs` + +### Added — New Requirements +- `req(quick-337)`: add **PERF-03** — fast-path initial diagnostic performance target +- `req(quick-348)`: add **SOLVE-22** — autoresearch-style iteration for formal model refinement +- `req(quick-350)`: add **SOLVE-23** — autoresearch-style iteration for solution-simulation-loop + +### Changed +- `feat(phase-53)`: solve-remediate b_to_f rewired from `/nf:model-driven-fix` to `/nf:debug` dispatch ### Deprecated -- **`/nf:model-driven-fix`** — Replaced with deprecation shim directing to `/nf:debug` +- **`/nf:model-driven-fix`** — replaced with deprecation shim directing to `/nf:debug` ### Removed -- **debug-formal-context.cjs single-shot call** — Replaced by 4-step formal pipeline in debug skill +- **debug-formal-context.cjs single-shot call** — replaced by 4-step formal pipeline in debug skill + +### Fixed +- `fix(formal)`: reduce NFHazardModelMerge state space from ~8T to 29K states +- `fix(quorum)`: sync FALLBACK-01 checkpoint to reference doc — prevent 1/1 consensus short-circuit + +### Tested +- UAT: Phase 50 (10 passed), Phase 51 (10 passed), Phase 52 (8 passed), Phase 53 (4 passed) +- `test(50-02)`: validate end-to-end variable flow consistency +- `test(quick-341)`: add nonce tests + orchestrator/reference doc updates +- `test(quick-350)`: add 9 tests for onTweakFix, rollback, TSV, when-stuck behaviors ## [0.40.2] - 2026-03-20 — Prerelease (next channel) @@ -70,19 +148,27 @@ See [0.40.1] for full changelog. This prerelease packages the same changes for t ## [0.40.1] - 2026-03-20 — Structural Enforcement & TLC Failure Classification -### Added -- **nf-scope-guard hook** — PreToolUse hook warns on out-of-scope file edits during phase execution -- **Context injection blocks** — 3 new context blocks in `nf-prompt.js` for richer session intelligence -- **Root cause quorum vote** — Step 0f added to `solve-diagnose` workflow for consensus-based root cause selection -- **TLC failure classifier** — 6-class pattern matching engine for TLA+ model checker output (`bin/classify-tlc-failure.cjs`) -- **F→C remediation dispatch** — classifier wired into `solve-remediate` formal→code layer with schema extension -- **FSM candidate detection** — phase researcher now identifies implicit finite state machines in target code -- **21 requirements from triage** — D→R and T→R requirement creation via `/nf:resolve`, linked to `quorum.pm` -- **CLASS-03 requirement** — formal requirement for TLC failure classification coverage +Version bump within the 0.40 milestone. Never independently tagged — rolled into [0.40.2] prerelease and then [0.41.0]. + +### Added — v0.40 Milestone (tagged v0.40) +- `feat(v0.40-03-01)`: wire nf-scope-guard hook and register in installer — PreToolUse hook warns on out-of-scope file edits during phase execution +- `feat(v0.40-01-01)`: add three context injection blocks to `nf-prompt.js` for richer session intelligence +- `feat(v0.40-02-02)`: add Step 0f root cause quorum vote to `solve-diagnose.md` +- `feat(resolve)`: create 21 requirements from D→R/T→R triage + link `quorum.pm` +- `req(quick-334)`: add **QUORUM-04** +- `req(quick-335)`: add **AGENT-04** + +### Added — Post-v0.40 (quick-336, shipped in 0.40.2-rc.1) +- `feat(quick-336)`: add TLC failure classifier with 6-class pattern matching engine for TLA+ model checker output (`bin/classify-tlc-failure.cjs`) +- `feat(quick-336)`: wire classifier into solve-remediate F→C dispatch + extend schema +- `req(quick-336)`: add **CLASS-03** — formal requirement for TLC failure classification coverage ### Fixed -- **Prerelease tag handling** — release workflow skips `v*-rc*` / `v*-next*` tags instead of erroring -- **TLA+ MCConvergenceTest** — resolved inconclusive formal check in convergence model +- `fix(ci)`: skip prerelease tags in release workflow instead of erroring +- `fix(formal)`: resolve `tla:mcconvergencetest` inconclusive check +- `fix(ci)`: skip version stamp when package.json already matches tag +- `fix(lint)`: use portable require path for `classify-tlc-failure` +- `fix(test)`: allow prerelease suffixes in CLI version output test ## [0.39.0] - 2026-03-19 — Dual-Cycle Formal Reasoning & CI/CD Formalization diff --git a/bin/nForma.cjs b/bin/nForma.cjs index a5ad1f9d18..aad21afc4d 100644 --- a/bin/nForma.cjs +++ b/bin/nForma.cjs @@ -2030,9 +2030,7 @@ async function checkHealthSingle() { lines.push(` Status: ${status}`); // Show provider info from providers.json or MCP env - let pdata; - try { pdata = readProvidersJson(); } catch { pdata = { providers: [] }; } - const providerMeta = (pdata.providers || []).find(pr => pr.name === slotName); + const providerMeta = providerMap.get(slotName); const displayProvider = env.ANTHROPIC_BASE_URL || (providerMeta && providerMeta.display_provider) || 'subprocess'; diff --git a/docs/assets/terminal.svg b/docs/assets/terminal.svg index 5980f97bc7..f0fda30375 100644 --- a/docs/assets/terminal.svg +++ b/docs/assets/terminal.svg @@ -80,7 +80,7 @@ - nForma — Consensus before code. Proof before production. v0.41.2 + nForma — Consensus before code. Proof before production. v0.41.3 Built on GSD-CC by TÂCHES. The task of leadership is to create an alignment of strengths so strong that it makes the system’s weaknesses irrelevant. diff --git a/package.json b/package.json index fd8cb5feff..698e30af06 100644 --- a/package.json +++ b/package.json @@ -1,6 +1,6 @@ { "name": "@nforma.ai/nforma", - "version": "0.41.2", + "version": "0.41.3", "description": "nForma — Multi-agent coding orchestrator with quorum consensus and formal verification (TLA+, Alloy, PRISM). Consensus before code, proof before production.", "bin": { "nforma": "bin/nforma-cli.js",