Fix Skill

Goal

Close the bug loop fast: capture evidence -> reproduce -> fix root cause -> verify -> watch for recurrence.

Modes

• •log-first: Start from live error data and incident reports.
• •repro-first: Start from automated reproduction when logs are weak or missing.

Choose log-first when 500s/exceptions are present. Choose repro-first for race conditions, stale UI state, click dead zones, and timing bugs.

Repo-Specific Inputs (TheRxSpot)

• •Live diagnostics endpoint: auth/api/live-errors.php
• •Client error ingest: auth/api/log-client-error.php
• •Structured diagnostics log: auth/logs/live-errors-YYYY-MM-DD.jsonl
• •Auth bootstrap correlation: auth/auth-helper.php (X-Request-Id)
• •Additional reference: references/therxspot-fix-loop.md

Safety Rules

• •Keep diagnostics gated:
  • •RXSPOT_LIVE_ERROR_API=1 only during active triage windows.
  • •Prefer header token auth (Authorization: Bearer ...), never query-string tokens.
• •Never modify runtime data files directly (auth/data/app.db, .db_initialized, sessions).
• •Never expose internals in client responses; keep details in logs only.

Workflow

1. •Confirm symptom and pick mode (log-first or repro-first).
2. •Acquire evidence.

• •log-first:
  • •Query recent events from auth/api/live-errors.php with filters (limit, lookback_minutes, search, sources).
  • •Group by signature: endpoint + message + stack fragment.
• •repro-first:
  • •Run deterministic probes (API replay or Playwright interaction loops).
  • •Capture request IDs, console errors, network failures, and timing windows.

1. •Build minimal repro.

• •For API failures: replay exact method/path/headers/body.
• •For UI race failures: script rapid navigation + repeated click/open/close interactions with small randomized delays.

1. •Fix root cause minimally.

• •Prefer lifecycle-safe init patterns for SPA/bfcache/turbo navigation.
• •Make handlers idempotent; prevent duplicate listeners; guard stale async responses.
• •Add contextual error logging with request/user/route context.

1. •Verify.

• •Required gates:
  • •npm run lint
  • •npm run test:php
• •If UI/race issue: rerun targeted Playwright scenario multiple iterations (>=20).
• •Re-query live diagnostics and confirm the failing signature drops or disappears.

1. •Report.

• •Include changed files, root cause, verification evidence, and residual risk.

TheRxSpot Command Patterns

• •Pull live errors:

curl -H "Authorization: Bearer $RXSPOT_LIVE_ERROR_API_TOKEN" \
  "https://therxspot.com/auth/api/live-errors.php?limit=200&lookback_minutes=180&sources=live,php,client"

• •Submit manual incident:

curl -X POST \
  -H "Authorization: Bearer $RXSPOT_LIVE_ERROR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"message":"notifications not marking read","severity":"error","type":"manual_report","steps":["open bell","click item"],"actual":"stays unread"}' \
  "https://therxspot.com/auth/api/live-errors.php"

• •Verification:

npm run lint
npm run test:php

Race-Condition Playbook

Use this when symptoms are "works after refresh" or "buttons dead until reload":

• •Re-initialize page logic on multiple lifecycles (DOMContentLoaded, SPA navigation events, pageshow).
• •Ensure init is idempotent and abort/replace old listeners.
• •Add short-lived diagnostics around event binding and async state transitions.
• •Run repeated navigation/interactions with jitter and collect failures by step index.

Autonomous Long-Run Loop

For unattended stabilization sessions:

1. •Poll live diagnostics every N minutes.
2. •Rank top recurring signatures by frequency + recency.
3. •Attempt one fix at a time (highest-impact first).
4. •Run verification gates and targeted repro iterations.
5. •Re-poll for regression.
6. •Stop when no critical signatures remain for a defined window.

Keep one fix per loop iteration to avoid confounding variables.

Completion Standard

Do not claim fixed unless:

• •Repro exists (or historical signature is clearly captured),
• •Root cause is identified,
• •Verification passes,
• •Post-fix diagnostics show improvement.