@agenit/cli 2.1.0 → 2.1.2

package/CHANGELOG.md

@@ -3,6 +3,155 @@
 All notable changes to `@agenit/cli` are documented here. The project
 follows [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [2.1.2] - 2026-05-17
+
+_Auto-generated entry. Edit CHANGELOG.md directly before
+the next release to curate this section._
+
+### Changed
+
+- Merge pull request #141 from muhammed-eldabea/claude/fix-release-workflow-Oasku (033a118)
+- fix(release): avoid SIGPIPE (141) in CHANGELOG generation (c528027)
+- Merge pull request #140 from muhammed-eldabea/FixSpecKitlaund (ef7473d)
+- feat(repl): less/vim-style pager — m/l/g/G keys + scroll-within-bubble (c66c172)
+- feat(repl): per-line pager + landing splash with hacker character (0a64cd1)
+- fix(repl): resolve Ctrl+D shortcut conflict (debug vs catch-up) (678701f)
+- feat(repl): flat Claude/Gemini-style chat layout — avatar in left gutter (9253725)
+- fix(repl): streaming bubble caps to tail-view instead of growing past viewport (04722d4)
+- perf(repl): memoise chat-scroll cull + EntryView to kill flicker (836d73b)
+- feat(repl): collapse long chat messages with Ctrl-E to expand (8e249dd)
+- fix(repl): goal pipeline animates only when a runner is attached (c7dcb61)
+- fix(repl): chat-input overlap + /goal resume runner spawn + sprite row accounting (18b89dd)
+- fix(orchestrator): inline speckit pipeline streaming in chat scroll (15df21a)
+- Merge pull request #136 from muhammed-eldabea/audit-followups (16daecb)
+- Merge pull request #139 from muhammed-eldabea/pixel-characters (a3f08c3)
+- Merge pull request #138 from muhammed-eldabea/feat/agile-opa-qa-ceremonies-radiators (645a2c2)
+- feat(pixel-characters): phase-transition wave/wake animation (5045145)
+- fix(test): make burndown upsert test deterministic in CI (369b17d)
+- feat(vscode-extension): per-helper pixel-character avatars in Squad tree (cb4eb85)
+- feat(vscode-extension): per-agent + per-skill avatars in Customizations panel (77f89ee)
+- feat(agile): information radiators — burndown writer, TUI health widget, /agile digest & burndown (f5db0dc)
+- feat(agile): close all 10 minor gaps — blocker detection, PM budget, OPA templates, QA DoD, ceremony gate modal, change requests, PMO standards, speckit queue (34d3e63)
+- feat(vscode-extension): chat avatar SVG + tool-bounce + vmodelTree per-stage sprite (d07cb8c)
+- feat(orchestrator): JobRow + GoalProgress sprites + V-Model stage characters (9b4ea6c)
+- feat(orchestrator): useFrameLoop hook + chat avatar + as:<role> directive (f0a72be)
+- test: add trigger-handler and human-relay unit tests + AgDR-0009 for PMO persona (18d6a85)
+- feat(sprites): fill remaining V-Model + persona + stakeholder ASCII art (d5a1f63)
+- feat(sprites): emit-webview build step → catalog.json + per-char SVG (835f28e)
+- feat(sprites): new @flow/sprites package + types + 2 proof characters (ba7634f)
+- fix: record velocity at retrospective + robust PMO QA failure detection (5b23ac4)
+- feat: wire ceremony trigger→persona actions + LLM AC evaluation in /qa (eb9934e)
+- feat: wire ResumeStrategyModal into App.tsx (#95/#114) (c76ca6d)
+- test: add unit tests for CeremonyEngine and OPA memory injection (a198a0c)
+- feat(goal): checkpoint cadence control via checkpoint_every_turns (audit #127 mech 2 follow-up) (f51145f)
+- fix: complete ceremony-worker ↔ goal-runner integration + opa.ts path (8e82e45)
+- feat(goal): /goal start --interactive + per-tick checkpoint API (audit #134) (1491bce)
+- feat(migrate): /migrate gemini-plan — recover raw Gemini plan-mode artefacts (audit #132) (3ce16e9)
+- feat(traceability): profile-aware ID + language defaults (audit #131) (5fa883f)
+- fix: wire Agile/OPA implementation gaps identified in post-commit review (53d4021)
+- feat(memory,goal): scope-relative SONA reward + live cost surfacing (audit #130, #133) (91db0b5)
+- feat(goal): scope-vs-spec complexity-band linter (audit #128) (6c4c758)
+- feat(session_start): mandatory memory-tool directives (audit #129) (bb968f5)
+- feat(goal): polish-task verifier — block COMPLETED ticks without evidence (audit #127, mechanism 1) (dfd79ae)
+- fix(goal): auto-index codedigest on goal start when missing (audit #125) (8a0a370)
+- fix(memory): extract atomic lessons before reasoning-bank append (audit #124) (5438195)
+- fix(session_start): profile-aware engineering trailer (audit #123) (b933ccd)
+- feat(agile+opa+qa): implement Agile persona engine, OPA, /qa, and session resume (#99-#114) (8d50ff8)
+- fix(audit): file-level paths in gitChangedFiles + extend isTestPath (b3e2f04)
+- fix(goal): parse COMPLETED: marker even when glued to preceding text (56124d4)
+- Merge pull request #118 from muhammed-eldabea/cancel-harden-and-features (97aaf80)
+
+## [Unreleased]
+
+### Added — Debug Mode (LangSmith-style traces)
+
+- `/debug` slash command — `on`/`off`/`status`/`pause`/`resume`/`release`/`skip`/`abort`/`recent`/`trace`/`replay`/`tag`/`config`/`deltas`/`thinking`/`redact`/`save`/`help`. All subcommands registered in autocomplete (`SUBCOMMAND_HINTS`).
+- `Ctrl+D` TUI hotkey toggles debug mode without typing.
+- Singleton trio: `debug-mode.ts` (live state), `run-registry.ts` (in-memory tree of runs persisted as `run.started`/`run.ended` events), `run-tracker.ts` (`withRun(args, fn)` helper + AsyncLocalStorage parent propagation), `debug-gate.ts` (Promise-based pause gates keyed by `runId`).
+- New `RunKind` taxonomy: `chain`/`llm`/`tool`/`agent`/`stage`/`retriever`/`parser`/`prompt`. Each run carries full inputs/outputs/tokens/cost/tags/metadata + a bounded `events[]` of streaming chunks.
+- Secret redactor (`packages/memory/src/redactor.ts`) with built-in patterns for Bearer / `sk-*` / `ghp_*` / `AKIA*` / Slack / Google API keys / password|token|api_key assignments. Recursive JSON walker; opt-out via `[debug] no_redaction = true`; extend via `extra_redact_patterns`.
+- TUI Debug widget (right rail) with 3-section expanded view: trace list / tree / detail. JSON syntax highlighting (cyan keys, green strings, yellow numbers, violet booleans/null) + word-wrap to panel width on INPUTS/OUTPUTS.
+- Web UI **Debug tab** in the Preact SPA — 3-pane LangSmith-style layout (trace list / tree / detail). Live polling via `/api/debug/state`. Config drawer for `enabled`/`persistDeltas`/`persistThinking`/`noRedaction`/`pauseBeforeKinds`. Pending-gate banner with Release/Skip/Abort buttons.
+- 9 new HTTP routes under `/api/debug/`: `state`, `traces`, `traces/:id`, `runs/:id`, `gates`, `gate/:id`, `config`, `replay/:id`, `runs/:id/tags`. Token-gated (writes require control token; viewer token returns 403).
+- **Tool replay** — every tool dispatch now captures full args + raw output via the extended `ToolDispatchEvent`. SPA's tool-run detail shows an editable JSON args textarea with **"Try call again"** button. Replay lands as a sibling run with `tags: ["replay"]` and `metadata.replayedFrom = <originalRunId>`.
+- LLM call wrap captures `{messages, response}` per run via `extractOutputs`. Chain wrap exposes the final assistant reply on the chat-turn row.
+- Streaming token deltas + thoughts-token markers — when debug mode is on AND `persistDeltas`/`persistThinking` is true, the chat path switches to `chatStreamJson` and emits `llm.delta` / `llm.thinking` RunEvents.
+- Two new `MemoryEvent` variants (`run.started`, `run.ended`) — persisted to `events.jsonl`; broadcast over `/ws/events`.
+- Five new Vitest files: `redactor.test.ts` (12), `debug-mode.test.ts` (7), `run-registry.test.ts` (13), `run-tracker.test.ts` (9), `debug-gate.test.ts` (12) — 53 new tests total. Full suite stays at 1203 passing.
+
+### Added — Harnesses widget
+
+- Single TUI surface for adding the four artifact types: **Skills** (`.gemini/skills/<name>/SKILL.md`), **MCP servers** (merged into `.mcp.json`), **Agents** (block appended to `AGENTS.md`), **Tools** (`.flow/tools/<name>.py`, chmod 0755).
+- Two sources for the harness body: **empty template** (one per kind) or **scaffolding agent** (one-shot LLM call with per-kind locked output format + `cleanScaffolderOutput` to strip code fences and conversational preambles).
+- In-TUI **vim editor** (`vim-editor-state.ts` reducer + `vim-editor.tsx` renderer). Subset of vim covered: normal/insert/command modes, `hjkl/arrows`, `0/$/gg/G/w/b`, `i/a/o/O`, `x/dd/yy/p/P`, `u` single-step undo, `:w/:q/:q!/:wq/:x`. Vitest reducer coverage (24 cases).
+- `harness-store.ts` singleton state machine with 11 modes. Drafts persist to `~/.agenit/harnesses/drafts.json` debounced 300ms.
+- `harness-installer.ts` with atomic tmp+rename writes, MCP JSON deep-merge, AGENTS.md block de-duplication. Name sanitization.
+
+### Added — Widget navigation, Agents dashboard, Recipes
+
+- `Ctrl+W` toggle widget nav mode. `←→` switches rails, `↑↓` cycles widgets within a rail, `Enter` expands, `Esc` collapses then exits. `1`–`9` quick-jump. Action hotkeys (single letters) when a widget is expanded.
+- `widget-registry.ts` + `use-widget-focus.tsx` — singleton store (not Context — App.tsx's `useInput` runs before any provider could wrap).
+- Legacy `Ctrl+T`/`Ctrl+]`/`Ctrl+G` shortcuts preserved and routed through the new expand mechanism.
+- Right-rail **AgentsWidget** with live activity (`reading X`, `running Y`, `asking model`) + per-agent token-cost sparkline + session spend total. Powered by `agent-activity.ts` (subscribes to `tuiBus` events) and `cost-to-agent-bridge.ts` (translates `cost` log events into `agent.cost` emissions).
+- Right-rail **RecipesWidget**. TOML-defined cross-widget orchestration loaded from `config/recipes.toml` (workspace) + `~/.agenit/recipes.toml` (user). Triggers: `hw.probe.failed`/`.ok`, `goal.tick.failed`/`.complete`, `agent.failed`/`.idle`, `tasks.changed`, `widget.action`. Actions: `submit-slash`, `system-message`, `wait N`, `tasks.assign`/`tasks.comment`/`tasks.priority`/`tasks.add`, `log`. Mustache-light template substitution. Per-recipe `guard.max_per_hour` + `guard.cooldown_seconds`.
+- Interactive Tasks expanded view (per-task menu — toggle done, cycle priority, add comment, assign worker from squad persona list, delete with confirm, add new task). Persistence via `> assignee:` / `> comment:` sub-bullets in `tasks.md` so the AI sees the annotations on next read. AI-suggestion sub-mode (`s` action) — backed by `task-curator.ts` one-shot LLM call.
+- Interactive HW Status expanded view (per-probe menu — reread, configure, info — for J-Link / CAN / Saleae).
+
+### Added — `/serve` slash command
+
+- `/serve start|stop|status|open` runs the full control SPA inside the running REPL process. Reuses `cmdServe` so all the orchestrator's dependency setup (broker, gemini, session, command registry, debug singletons) is shared. The Web UI Debug tab sees the live run tree from the same REPL.
+- Distinct from the legacy `/web` command which only starts the read-only state-mirror viewer.
+- Subcommands registered in autocomplete picker (`SUBCOMMAND_HINTS`).
+
+### Added — `[debug]` config section + `[experimental] widget_nav`
+
+- New `DebugSection` in `agenit.toml`: `enabled` (master), `persist_deltas`, `persist_thinking`, `pause_before_kinds`, `max_persisted_bytes`, `max_events_per_run`, `retention_runs`, `no_redaction`, `extra_redact_patterns`, `persist_trace_files`. Slash command mutates the in-memory singleton; `agenit.toml` only changes when the user runs `/debug save`.
+- `[experimental] widget_nav` flag (default `true`) — hard kill switch for the widget navigation system.
+
+### Changed
+
+- Documentation folder consolidated: `doc/` → `docs/`. All `doc/<path>` references updated across the codebase (source, tests, config, skills). Three new files added: `docs/debug-mode.md`, `docs/harnesses.md`, `docs/widget-navigation.md`. `docs/README.md` index extended.
+- Root `README.md` extended with a **What's new** section and entries in the Documentation table for the new features.
+- `ToolDispatchEvent` (gemini-client) — `started` events now carry full `args` (was: summary only); `completed` events now carry full `output` (was: 200-char preview only). Backwards-compatible — fields are optional and the legacy `argSummary`/`outputPreview` remain.
+- `tool-executors.ts` dispatcher observer extended to populate the `runRegistry` so every tool call becomes a child run of the LLM run that invoked it.
+- Subcommand autocomplete entries added for the slash commands that were missing them: `/debug`, `/serve`, `/web`, `/harnesses`. Typing `/<cmd>` + TAB now lists all subcommands instead of silently doing nothing.
+
+### Fixed
+
+- Debug widget detail pane previously truncated long response strings with `wrap="truncate-end"`, causing them to bleed off the right edge of the panel. Now uses `wrap="wrap"` and a per-line JSON syntax colorizer so multi-line redacted outputs are readable and word-wrapped to the available panel width.
+- Run-registry's `RunStatus` was being persisted as `running` on `run.ended` events when the inner function returned. Now narrowed to the terminal-only union (`ok` / `error` / `aborted` / `skipped`) at the persistence boundary.
+- `String.prototype.replace` callback signature gotcha in `redactor.ts` — for regexes without capture groups, the second arg is the *offset* (a number), not undefined. Patterns with no capture group were silently failing to redact because the heuristic check `if (group1 !== undefined)` treated the numeric offset as truthy. Fixed by inspecting the pattern source for non-`(?:` parentheses to decide whether the regex declares a capture group.
+
+## [2.1.1] - 2026-05-14
+
+_Auto-generated entry. Edit CHANGELOG.md directly before
+the next release to curate this section._
+
+### Changed
+
+- Merge pull request #98 from muhammed-eldabea/TestMain (4ccb158)
+- feat(gemini-client): 429 fallback, request timeout, crash detection + warm session_start (7907845)
+- feat(perf): hermetic + live LLM response-time benchmarks (704d9a6)
+- fix(test): canonicalize tempdir in test_lsp_tool for macOS (80589ab)
+- Merge pull request #94 from muhammed-eldabea/claude/ai-dev-2026-analysis-CnFOW (b9829ef)
+- fix(security): clear CodeQL alert + sweep docs to current state (85d399a)
+- fix: close four self-raised follow-up gaps from prior commits (68c7ea4)
+- docs(agent): make code/memory tools self-discoverable + warm the stance (a37502f)
+- feat(codedigest): semantic + structural retrieval for internal agents (77c7fa2)
+- feat(codedigest): Crabviz-style LSP call graph (DKB follow-up to #82) (5fd7dff)
+- feat(cli #92): /pr-autofix — Critic + apply deterministic fixes on PR (746e295)
+- feat(observability #91): /pr-metrics AI-vs-human cohort report (61333bb)
+- feat(mcp #90): hardware tools (JLink/CAN/LIN/Saleae/DBC) over MCP (0ee4e7b)
+- feat(cli #87): /diff-walk — Critic per hunk (5016806)
+- feat(speckit #88): /speckit critique subcommand (5cafdab)
+- feat(codedigest #82): bidirectional symbol graph + tree-sitter AST scaffold (32b969b)
+- feat(squad #86): swe6-critic skill + opt-in Critic phase (65be70f)
+- feat(budgets #93): per-command + per-session token-budget ceilings (d464215)
+- feat(security #89): --air-gapped runtime mode (83030d6)
+- feat(security #84): MCP server pinning + skill audit scanner (faf2832)
+- feat(security #83): slopsquatting / phantom-package defense hook (50775ed)
+- docs(#85): rewrite CLAUDE.md to reflect TypeScript monorepo + drift gate (fc46f91)
+
 ## [2.1.0] - 2026-05-13
 
 _Auto-generated entry. Edit CHANGELOG.md directly before