The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes node scripts/detect-scope.js and various git commands (git rev-parse, git status, git diff, git log) to identify development changes within the repository. These operations are restricted to the local environment and are used solely for scope detection.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local git repository and user input, which constitutes an attack surface for indirect prompt injection.
Ingestion points: Reads repository state (file names, diffs, logs) via scripts/detect-scope.js and collects user questions via the AskUserQuestion tool.
Boundary markers: The skill relies on its sub-skills (/recap-doc and /recap-ask) to provide necessary isolation and instruction-override protection for the ingested data.
Capability inventory: The workflow utilizes Bash(node:*), Bash(git:*), Write, Skill, and mcp__codex__codex capabilities.
Sanitization: Integrates scripts/security-redact.js to filter and redact sensitive information from the generated recap documentation and subsequent Q&A context.
[SAFE]: The skill follows security best practices by resolving working directories to the repository root, using ephemeral temporary storage for intermediate data, and strictly prohibiting git mutation operations (e.g., commit, push, reset).

post-dev-recap