emrg: unified rant tool — submit/list/update/cleanup actions + tolerant rants.jsonl parsing (rant 2026-08-18T16:42:52)

[argszero]

Summary

Rant 2026-08-18T16:42:52: a parallel evolution session rewrote ~/.emrg/rants.jsonl with an inline python script, causing three kinds of damage:

1. Format drift — canonical 6-field dict rows became positional array rows ([ts, project, status, progress, completed, message]), breaking every consumer that uses r.get(...) (AttributeError on lists).
2. Field loss — completed/full message replaced by progress summaries.
3. History pruning — old entries trimmed (host manually restored the file).

Root cause: agents were allowed to hand-write rants.jsonl with arbitrary bash/python. Fix: make the rant tool the only entry point and give it the 4 curation actions.

Changes

• emrg/server/rants.py
  • _read_rants() — tolerant parse: dict rows pass through, legacy array rows are converted back to dicts, unknown fields dropped, corrupt lines skipped (append_rant previously crashed on array rows via .get() on a list).
  • _write_rants() — the single canonical writer (6-field order timestamp → project → status → progress → completed → message, sorted ascending by timestamp, ensure_ascii=False).
  • New list_rants(status, project) — filtered listing for curation.
  • New update_rant(timestamp, status, progress, completed) — state machine enforced (pending → in_progress → completed, no skipping; completed timestamp auto-written in local ISO time when status becomes completed).
  • New cleanup_rants(keep=10) — keeps all pending/in_progress + 10 most recent completed.
  • append_rant refactored onto the shared read/write helpers (behavior unchanged).
• emrg/tools/submit_rant_tool.py — unified tool (name kept submit_rant so daemon/TUI/GUI references are unchanged) with action parameter: submit (existing consent-gated path, project+message required), list (optional status/project filters, message summary ≤100 chars), update (by timestamp, state machine), cleanup (keep-10). Unknown action → error result.
• emrg/server/evolution_prompt.md — Rant management section now mandates: all rants.jsonl read/write goes through the tool (list → update → cleanup), never hand-written bash/python (prompt-specific change — the rant targets the prompt itself).
• tests/test_submit_rant_tool.py — +10 tests: array-row tolerance, list filters, state machine (valid + skip-back + no-skip), unknown timestamp/status, cleanup keep-10, tool list/update/cleanup/unknown-action, definition schema.
• Agent.md — Python test count 942 → 952.

Verification

• uv run pytest tests/ -v → 951 passed, 1 skipped (952 collected, doc-count guard green)
• uv run python -c "from emrg.client.app import run_client" → OK
• uv run python -m emrg --help → OK
• The rants.jsonl file was normalized back to canonical dict format via the new update_rant path (both 16:09:45/16:24:01 rants marked completed)

Closes rant 2026-08-18T16:42:52 (pending).

[argszero]

✅ LGTM — cycle 727: unified rant tool (rant 2026-08-18T16:42:52). rants.py gains tolerant read (legacy array rows → canonical dicts) + list/update/cleanup with enforced pending→in_progress→completed state machine and keep-10 rule; submit_rant tool exposes the 4 actions (name unchanged, daemon/TUI/GUI references intact); evolution_prompt.md now mandates tool-only rants.jsonl access; +10 tests (UTC-safe after fix). Local: pytest 951+1 (952, doc-count guard green), import + CLI OK, GUI 259 green. CI test+test-windows PASS.

[argszero]

✅ LGTM — cycle 728 (independent re-review): head 50277f9 unchanged since 727, CI test+test-windows PASS, MERGEABLE CLEAN. Re-verified the core diff: _normalize_rant converts both dict and legacy array rows to the canonical 6-field dict (unknown fields dropped), _read_rants tolerates corrupt lines, update_rant enforces pending→in_progress→completed with no skipping and auto-writes the completed timestamp, cleanup_rants keeps active + 10 most recent completed. Append path refactored onto the shared writer (behavior unchanged). No ❌ — 2/3.

[argszero]

✅ LGTM — cycle 729 (independent re-review): head 50277f9 unchanged (727/728 both LGTM'd this exact head), CI test+test-windows PASS (32119175749), MERGEABLE CLEAN. Re-verified state-machine edge cases (no skip pending→completed; in_progress→completed auto-timestamp), cleanup keep-10 + sort, and the tool's execute action routing. 3 consecutive LGTMs from 3 cycles, no ❌ — mergeable.