session-hardener

name	session-hardener
description	Explicit post-session repo hardening review for Codex/AI coding sessions.
disable-model-invocation	true

Session Hardener

Turn the current AI coding session into exactly 3 durable repo-hardening recommendations. Recommend only; do not edit, stage, commit, archive, or apply fixes unless the user separately asks for implementation.

Workflow

1. Inspect the current repository before judging the session:

   • Read the nearest AGENTS.md.
   • Run read-only git inspection such as git status --short, git diff --stat, and focused diffs for touched files.
   • Read touched files, tests, scripts, or docs needed to understand behavior.

   Completion criterion: repo evidence covers the files or workflows that the session changed or struggled with.

2. Collect current session evidence:

   • Prefer scripts/collect_current_session.py --pretty.
   • If CODEX_THREAD_ID is unavailable, use scripts/collect_current_session.py --latest-for-cwd . --pretty and label it "latest session for cwd".
   • If collection fails, use live conversation context and state the gap.

3. Look for repo-hardening problems, not code-review nits:

   • repeated agent confusion or rediscovery
   • missing, stale, or ambiguous AGENTS.md rules
   • missing tests, fixtures, or verification commands
   • fragile scripts, config defaults, or bootstrap behavior
   • tooling gaps that caused avoidable approvals, failed commands, or manual checks

4. Pick exactly 3 problems by recurrence risk and usefulness of a durable repo update. Prefer actionable repo changes over process advice.

   Completion criterion: each selected problem has objective evidence and at least one concrete repository-level fix.

5. Return the response contract below. Do not include patches.

Evidence Rules

• Ground each problem in objective evidence: a short session quote, failed command excerpt, approval pattern, or clickable local file link.
• Quote only short snippets; redact secrets and tokens.
• Use file links like [AGENTS.md](/absolute/path/AGENTS.md:12).
• If evidence is thin, say so and set confidence to medium or low.
• Do not dump raw transcripts or long tool outputs.

Response Contract

Return exactly 3 numbered problems:

1. **Problem:** Paragraph explaining the recurring risk and why it matters.

   **Evidence:** Paragraph with short dialogue/tool quote or repo file link.

   **Fix 1:** Paragraph describing one durable repo update.

   **Fix 2:** Optional paragraph for a meaningful alternative or complement.

   **Confidence:** high|medium|low.

Collector Script

Run from this skill directory or pass the full path:

python3 config/codex/skills/session-hardener/scripts/collect_current_session.py --pretty

Useful options: --thread-id, --latest-for-cwd, --codex-home, --max-message-chars, --max-output-chars, and --max-events.

The script is read-only with respect to Codex state and repository files. It must not update state_5.sqlite, move rollout files, archive sessions, or write generated artifacts.