daimon 0.7.0 → 0.8.0

package/CHANGELOG.md

@@ -4,6 +4,63 @@ All notable changes to Daimon are documented here. The format follows [Keep a Ch
 
 ## [Unreleased]
 
+## [0.8.0] — 2026-05-20
+
+Strategic theme: **Mature daimon.** v0.5–v0.7 added enormous surface; v0.8 turns inward to polish the surface and harden the internals — a big CLI polish, a reliability pass, self-observability, and a plug-in surface for doctor rules, plus a first-class Tests page that finally lands part of the v0.7.1 dashboard-test debt (H3 contrast audit deferred again to v0.8.1).
+
+### Added (M45) — Tests dashboard + H1 carry-over
+
+- **T1 — Structured `test`-target summaries.** `parseTaskSummary` (in `src/taskRunner.ts`) now recognizes seven runners: Jest, Vitest, Karma, Playwright, pytest, RSpec, `cargo test`, and `go test` (the last sums multi-package `ok|FAIL pkg D.DDDs` lines). Output shape is now `{ passed, failed, total, suites?, durationMs?, framework, failedTests? }`. `failedTests` captures `{ name, file, line }` from the per-runner failure formatting and is the source of the dashboard's failed-test jumper. `task_runs.summary` continues to be a JSON blob — the new fields are additive.
+- **T2 — `Tests` dashboard route (`/tests`).** New lazy-loaded route. One expansion-panel per app with at least one `test`-named run in the last 30 days. Title row carries an OK / FAIL pill + `passed/total · N failed` label + framework + duration + "<N>m ago" + run count. Body shows a 30-row pass/fail trend ribbon (one tick per run) and, when the most-recent run failed, an inline list of failed tests.
+- **T3 — Failed-test jumper.** Each failed-test row that carries a `{ file, line? }` payload renders a clickable `vscode://file/<path>:<line>` link, identical scheme to the M22 errors panel.
+- **T4 — Nav + shortcut.** New `/tests` entry on the nav rail (`science` icon) and `g x` keyboard chord, alongside the existing `g t`/`g h`/etc. shortcuts.
+- **H1 (carry-over — third schedule slot, partial).** Vitest harness shipped at `dashboard/vitest.config.ts`. Five specs in `dashboard/src/app/tests-page.spec.ts` cover the pure signal-bearing logic — `parseSummary`, `vscodeUri`, `summaryLabel`, `pillKindFor` — extracted into `tests-page-helpers.ts` to keep them runtime-independent. `dashboard/package.json` `test` script is now `vitest run` (no `ng test --watch=false` blocker). The deeper per-component render coverage requires an Angular-linker preset (e.g. `@analogjs/vitest-angular`) — that preset has been moved to v0.9 to keep M45 inside one weekend. Playwright smoke similarly deferred. See `PLAN-v0.8.md` Status section for the trade-off.
+- **H3 — WCAG AA contrast audit.** Deferred to v0.8.1 per the locked descope path (`"If M45 grows, drop H3 first"`). M3 surface-tonal defaults from v0.7 still apply; the dedicated AA sweep across light + dark + reduced-motion remains a focused future pass.
+- **Test surface.** New `test/task-summary.test.mjs` asserts the parser for jest / vitest / pytest / rspec / cargo / go / playwright / null. `dashboard/src/app/tests-page.spec.ts` exercises the five tests-page helpers under Vitest.
+
+### Added (M44) — Plug-in surface for doctor rules
+
+- **P1 — Plug-in loader.** New `src/plugins.ts`. On daemon start, daimon scans `~/.daimon/plugins/doctor-*.mjs` (configurable via `plugins.dir` in `daimon.config.json`). Each file is dynamically imported with a cache-busting query string so plug-ins can be hot-edited and reloaded by restarting the daemon. Files that fail to import (or fail shape validation) are logged as `[daimon] plug-in skipped: <file> — <reason>` and never crash the daemon.
+- **P2 — Typed `DoctorPlugin` interface.** Defaults to a read-only `scan(ctx)`. Optional `fix(finding, ctx)` and `undo(finding, ctx)`. `DoctorContext` exposes `{ config, apps, history.querySelfMetrics, mutations }` — the `mutations` surface is intentionally narrow and re-uses the M36 mutation primitives (no shell-out, no package-manager calls, no edits to user source).
+- **P3 — Permission gating is fully opt-in.** No plug-in's `fix` ever runs unless its `name` is in `doctor.autoFix.permitted` — the same gate built-in rules use. This applies uniformly to the bundled sample (`src/templates/plugins/example-doctor.mjs`) and to any user-authored plug-in. The loader does **not** track plug-in origin; there is no "bundled / trusted" branch. The loader also rejects names that collide with built-in rules or with another plug-in already loaded in the same pass.
+- **P4 — `daimon plugin list|show <name>|validate <path>`.** New CLI subcommands. `list` returns installed plug-ins with `{name, description, file, status, error, findings}`. `show <name>` prints one plug-in's manifest + last-run findings. `validate <path>` sanity-checks a plug-in file *without* loading it into the running daemon (useful during development). New HTTP endpoints: `GET /api/plugins`, `POST /api/plugins/scan`.
+- **P5 — Sample plug-in template.** `src/templates/plugins/example-doctor.mjs` ships in the tarball (~1.8 KB). Demonstrates a complete `scan` + `fix` cycle with the one-line opt-in instruction. Used as the on-ramp for the permission model — the comment block documents the exact `permitted: [..., 'example-doctor']` line a user must add to enable the `fix`.
+- **P6 — Dashboard surfacing.** Doctor page gains a "Custom rules" section. One card per plug-in with a status chip (`ok` / `failed`), description, file path, error detail (when failed), and per-finding rows with severity chips. Errors surface in the card header. Reuses existing M3 token styling — no new tokens.
+- **P7 — Skill text update.** `~/.claude/skills/daimon/SKILL.md` now lists `daimon plugin list|show|validate`, `daimon self`, `daimon doctor --self`, and `daimon completion` so agents can document a workspace's plug-in surface when asked.
+- **Decisions locked in.** Plug-in surface in v0.8 is **doctor-only** — parser-hint and event-hook plug-ins are deferred to v0.9+ until real-world plug-ins inform the right shape.
+- **Test surface.** New `test/plugins.test.mjs` asserts (1) only `doctor-*.mjs` files are picked up, (2) malformed plug-ins are reported as failed without crashing the loader, (3) names that collide with built-ins are rejected, (4) `runPluginScans` populates `lastFindings` and survives a throwing scan by quarantining the plug-in, and (5) `validatePluginFile` works standalone.
+
+### Added (M43) — Self-observability
+
+- **O1 — `GET /api/self`.** New endpoint returning daimon's own metrics: `{ pid, version, uptimeMs, rssMB, heapUsedMB, heapTotalMB, eventLoopLagMs, eventLoopLagP95Ms, historyDbQueryMs:{p50,p95,p99}, lockContentionCount, tickIntervalMs, lastTickAt }`. Backed by a new `SelfMetricsCollector` (in `src/selfMetrics.ts`) that probes event-loop lag with a 1s `setInterval` delta and rolls a 60-sample sliding window. `lockContentionCount` and the `historyDbQueryMs` percentiles are wired through call-site instrumentation (additive).
+- **O2 — `daimon doctor --self`.** New `--self` flag on `daimon doctor` runs self-checks: heap above 256 MB, event-loop lag p95 above 100 ms, history-db query p95 above 50 ms. Reports findings in the same `{ok, checks, metrics}` shape as the existing F58 doctor rules. Read-only by design (no auto-fix — self issues require a daemon restart or config tuning).
+- **O3 — `self_metrics` table in history.db.** New `self_metrics(id, ts, rssMB, heapUsedMB, eventLoopLagMs, historyQueryP95Ms)` table created by the existing `CREATE TABLE IF NOT EXISTS` migration — v0.7 databases upgrade in place. The daemon persists one row per minute via the new `History.recordSelfMetric`. Retention follows the existing `history.retentionDays` policy. `daimon snapshot <name>` carries the last 60 rows as `payload.selfMetrics` for diagnostic context.
+- **O4 — Self chart on the Trends dashboard route.** New "Apps / Self" toggle on `/trends`. When **Self** is selected, an additional `dm-trend-chart` card renders rssMB / heapUsedMB / eventLoopLagMs as line series from `/api/self/history`. Reuses the existing chart.js lazy chunk — initial-route gzip stays at **126.46 KB** (v0.7 baseline 126.45 KB, ceiling 130 KB).
+- **O5 — Self-warn events.** When the event-loop lag exceeds 100 ms for ≥5 consecutive ticks, `SelfMetricsCollector` emits a synthetic `{ app:'__daemon__', type:'self-warn', message:'event loop lag sustained: <N>ms (<K> consecutive ticks)' }` event. Surfaces on the existing Events feed alongside per-app events. New `'self-warn'` value in `AppEventType`; the loader only re-fires on rising-edge transitions to avoid flooding.
+- **Test surface.** New `test/self-metrics.test.mjs` asserts (1) `snapshot()` returns plausible pid/version/rss/heap/uptime/lag numbers, (2) `recordQueryMs` correctly feeds the p50/p95/p99 percentile estimates, and (3) the self-warn setter is wired without throwing.
+
+### Added (M42) — Reliability pass
+
+- **F1 — Parser fuzz tests.** New `test/parser-fuzz.test.mjs` feeds 2,000 deterministically-random mixed-ANSI / multibyte / lone-surrogate / multiline-fragment lines into `parseLine`. Asserts no exceptions, no >10ms slow tail above 2% of iterations, and total time <10s per run. Two regression cases cover empty / whitespace / ANSI-only input and lone surrogate halves.
+- **F2 — History.db stress test.** New `test/history-stress.test.mjs` inserts 100,000 events + 50,000 compile rows + 10,000 bundle rows into a temp sqlite (flushing in 5k-row batches via a new `History._flushForTest()` escape hatch), then queries `/api/history/trends` for every (metric × window × app) combination across three passes. Asserts trends p95 <50ms, p99 <200ms, db size <50MB.
+- **F4 — Lock-file contention.** New `test/lock-contention.test.mjs` isolates `~/.daimon` via temp `HOME`/`USERPROFILE`, then asserts (1) `readLock` tolerates missing/corrupt files, (2) it prunes stale locks pointing at dead PIDs, and (3) parallel `writeLock` calls leave exactly one parseable JSON lock — atomic rename never produces a partial payload.
+- **F5 — Error-map TTL.** New optional `errorRetention.maxAgeMs` config (default `86400000` — 24h) prunes `AppState.errors` entries older than `maxAgeMs` and not seen since. A new `Registry.pruneOldErrors()` walks every app's error map; the daemon wires it on an hourly tick in `main.ts`. Prevents unbounded growth in long-running daemons.
+- **F6 — Log buffer cap audit.** New regression in `test/reliability.test.mjs` exercises the per-app `logBuffer` with a 10,000-line burst and asserts the rolling window stays at `LOG_BUFFER_MAX` and contains exactly the latest 500 lines.
+- **F3 — Daemon crash auto-recovery soak (manual).** Documented in `test/SOAK.md` — 30-iteration kill-and-respawn loop with assertions on orphan PIDs and stale-lock pruning. Not run in CI (would dominate suite budget).
+- **F7 — Memory soak (manual).** Documented in `test/SOAK.md` — 24h procedure with 5 simulated apps, capturing baseline RSS via `daimon self` and the new M43 `self_metrics` table. Acceptance: <10% RSS growth over 24h.
+
+### Added (M41) — CLI polish
+
+- **C1 — Unified help system.** `daimon --help` now groups commands by category (lifecycle / queries / agent verbs / introspection / config / claude / plugin). Per-command help is available as `daimon <verb> --help` (or `daimon help <verb>`) and follows a single template — synopsis, description, options table, examples, exit codes — driven entirely by `cliSurface.ts` so help, completion, and skill text cannot drift.
+- **C2 — Shell completions.** New `daimon completion <bash|zsh|fish|powershell>` emits a completion script to stdout for the chosen shell. All four shells supported at launch. Completions cover verbs, aliases, and common flags; the bash/PowerShell scripts also auto-complete app names by querying a running daemon (`--no-spawn` is used to avoid starting one for completion).
+- **C3 — Color and TTY awareness.** New `--no-color` flag and `NO_COLOR` env var disable ANSI coloring. When stdout is **not** a TTY (the agent / piped case), output is **never** colored — compact JSON default from M26 is preserved unchanged. TTY mode colors errors red, hints dim, headings bold, keywords cyan. `FORCE_COLOR=1` honored as the per-CI override.
+- **C4 — Levenshtein "did-you-mean" suggestions.** Unknown commands and unknown apps now emit a `did you mean '<X>'?` hint. Tuned to a 2-edit-distance threshold to avoid noise. Apps lookup polls the running daemon.
+- **C5 — Flag convention block in `--help`.** A new "flag conventions" section in the main help documents the canonical flag vocabulary (`--timeout`, `--since`, `--budget`, `--until`, `--app`, `--profile`, `--full/--compact`, `--stream`, `--explain`, `--dry-run`, `--yes`). No flag renames — convergence is documented.
+- **C6 — Muscle-memory aliases.** New aliases: `daimon ls` → `list`, `daimon ps` → `status`, `daimon log` → `logs`. Aliases are documented in `--help` but not duplicated in the main verb list and do not conflict with existing verbs.
+- **C7 — Error message rewrite.** CLI errors now carry a `{ "error": "<short>", "hint": "<actionable next step>", "exit": <code> }` shape in JSON (piped) mode. TTY mode renders the one-line human variant (`error: …` + dim `hint: …`). The "daimon is not running" error nudges the user to `daimon daemon start --detach`. Unknown command / unknown app errors carry suggestions.
+- **C8 — `daimon --version` / `daimon --about`.** `--version` now prints the bare SemVer string (was `{"version":"..."}`). `--about` prints `{version, nodeVersion, platform, configPath, lockPath, running, runningPid, runningPort, claudeArtifacts}` — useful for bug reports. `daimon self` (new lightweight verb) reaches the daemon's `/api/self` endpoint introduced in M43.
+- **Test surface.** New `test/cli-surface.test.mjs` asserts every subcommand has the full help template, aliases resolve to canonical verbs, did-you-mean returns sensible candidates within 2 edits, and the completion script generator produces non-empty output for all four shells.
+
 ## [0.7.0] — 2026-05-20
 
 Strategic theme: **Memory and reach.** v0.7 makes the *past* actionable (history surfaces) and broadens *what* daimon can manage (polyglot dev servers, whole-workspace orchestration, refreshed TUI).