Claude Code / Codex / Cursor Agent Skill: defensive Dependabot triage with lockfile parity & exposure mapping. Plus team git workflow scripts (releases, PR helpers, branch resets).
Command Claude Code
Start-of-session rehydration. Reads NEXTSESSION.md + UPDATES.md tail + last 3 ADs, runs drift check on prior verification commands, proposes next move. See session-loop skill.
Command Claude Code
Mid-session light save. Overwrites NEXTSESSION.md with a minimal snapshot. Flags newly-crossed irreversible boundaries since last save. See session-loop skill.
Command Claude Code
Self-assess whether to compact context window at a task boundary, persisting carry-forward state to NEXTSESSION.md (or task metadata) before any compaction. See session-loop skill.
Command Claude Code
Append ADR-style decision to the project's ledger. Forces context/decision/alternatives/supersedes/blast-radius/reversibility fields. Append-only — never retro-edits. See session-loop skill.
Command Claude Code
Reality vs ledger reconcile. Re-runs verification commands and probes resources from the last 3 ADs. Surfaces concerning divergence. Read-only. See session-loop skill.
Command Claude Code
Scan unresolved threads across NEXTSESSION.md, UPDATES.md, ledger ADs without follow-up, in-repo TODOs, open PRs, and pending agent tasks. Ranks by urgency. Read-only. See session-loop skill.
Command Claude Code
End-of-session safety gate + handoff write. Refuses dirty stops; writes UPDATES.md, ledger entries, and NEXTSESSION.md. See session-loop skill.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: