webdevtodayjason Release Notes

fix(shell): claude-use / claude-whoami resolve v3 subctl explicitly

The shell helpers called bare subctl config show/accounts. Once the v4 subctl-chat-tui entry point at ~/.local/bin/subctl sorted ahead of ~/bin on the interactive PATH, bare subctl resolved to v4, which doesn't dispatch v3 verbs — so config show printed v4 help and the account parser found nothing. Result: claude-use listed no profiles, couldn't switch/select; claude-whoami always said "custom".

Fix: bake a _subctl_v3 resolver into the generated shell-aliases.sh and route all four bare-subctl call sites through it, bypassing PATH. Mirrors the existing claude-teams shim. The intentional bare-claude account guard is unchanged.

To pick up the fix in an existing shell: source ~/.zshrc or open a new terminal.

PR #48.

feat(fitness): dashboard fitness panel (Kernel Fitness Phase 3) (v3.3.1)

New Fitness tab in the dashboard sidebar (between Memory and Vault) exposing the metrics from Phase 1 (engagement ledger, v3.1.0) and Phase 2 (fitness writer, v3.3.0). Pure read-only observability — no daemon changes, no writes to either ledger.

Panels:

• Now: current stall composite + engagement rate with sparklines
• 7-day trend: inline SVG line chart (no Chart.js dep)
• Recent surfaces (24h): last 30 events with outcomes
• Health pill: red/yellow/green answer to "is Evy learning?"

New endpoints:

• GET /api/evy/fitness/ledger[?window=24h|7d|30d|Nh|Nd]
• GET /api/evy/engagement/ledger[?window=...][?type=...]
• GET /api/evy/fitness/health

All endpoints handle file-missing gracefully (empty array or red verdict, never 500).

Negative criterion preserved: dashboard reads ledgers via node:fs through dashboard/lib/fitness-api.ts; never imports from components/evy/fitness-writer.ts or components/evy/engagement-tracker.ts. Supervisor-prompt assembly path still has zero ledger visibility. fitness-ledger-isolation.test.ts + engagement-ledger-isolation.test.ts still pass (5/5).

Polling: 60s (writer emits hourly, faster is wasted).

After this ships: 30-day soak before Phase 4 (scaffold extraction = v3.3.2). Soak lets baseline fitness data accumulate before any optimization pressure is applied.

Tests: 196/196 dashboard tests pass (was 168 — added 28 fitness-api tests covering path resolution, window parsing, jsonl reading, both filtered readers, and computeHealth's 8 verdict scenarios).

Closes Phase 3 of Initiatives/Kernel Fitness — engagement + refiner + judge.md.

Co-Authored-By: Claude Opus 4.7 [email protected]

fix(fitness): CodeRabbit review feedback on v3.3.1

Three defensive fixes from CodeRabbit review of PR #33:

• Escape missing_data_reason in fitness.js — aligns with the escapeHtml pattern already used for surface_type elsewhere in the file. Defense in depth even though the source is our own writer.
• Filter unparseable timestamps before slope computation in computeHealth — Date.parse on a malformed window_start returns NaN, which silently poisons the linear-regression sums and gives a bogus verdict. Now we Number.isFinite-gate before mapping to slope points.
• Add Cache-Control: no-store to all three new observability endpoints — fitness data is per-window snapshots, must never be browser-cached or proxy-cached.

Skipped: CodeRabbit's "outside diff range" CHANGELOG dup at line 3138 — that block is pre-existing in main (origin/main also has the dup starting at line 3099). Cleanup belongs to a separate PR, not this feature.

Tests: 196/196 dashboard suite still passes.

Co-Authored-By: Claude Opus 4.7 [email protected]

Co-authored-by: Claude Opus 4.7 [email protected]