The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell utilities such as jq, grep, and bash to implement session hooks in the .claude/settings.json file. These hooks provide automated reminders and safety checks for the agent during the development lifecycle.
[COMMAND_EXECUTION]: Multiple Git commands are executed (git status, git ls-files, git remote) to audit the project's structure, identify tracked dependencies, and determine the appropriate CLI tools for the repository's hosting provider.
[SAFE]: The skill includes a defensive check for sensitive files (e.g., .env, *secret*) using git ls-files. This is designed to detect and warn the user if sensitive credentials have been accidentally committed to the Git history.
[SAFE]: The 'Memory Guard' functionality explicitly blocks agents from writing to local hidden directories (e.g., ~/.claude/projects/.../memory/). This is a security and reproducibility feature that ensures project state is stored in accessible, version-controlled locations.
[INDIRECT_PROMPT_INJECTION]: The skill processes repository metadata and configuration files to populate instructions for the AI agent.
Ingestion points: Manifest files (package.json, Cargo.toml), linter configurations, and CI workflow YAMLs in the target repository.
Boundary markers: The generated AGENTS.md file uses standard Markdown headers for structural organization; the template uses <...> placeholders for generated content.
Capability inventory: File writing, Git operations, and shell command execution via agent harness hooks.
Sanitization: Project-derived facts are subjected to a 'sanity-check' against the codebase, and user confirmation is requested for ambiguous project attributes.

repo-boost