The Agent Skills Directory

[DATA_EXFILTRATION]: The skill incorporates proactive security measures to prevent data leaks. It includes regex-based patterns to scan for and block common secrets (OpenAI, Anthropic, AWS, GitHub tokens, etc.) before commits are finalized.
Evidence: references/core.md and scripts/pre-commit-block-secrets.sh contain extensive pattern matching for credential detection.
[COMMAND_EXECUTION]: The skill utilizes local shell scripts (bump-manifests.sh, check-manifests.sh, install-hooks.sh) and standard CLI tools (git, gh) to perform repository management tasks.
Evidence: SKILL.md instructs the agent to invoke these scripts for version management and hook installation.
[PROMPT_INJECTION]: The skill processes repository content (manifests, READMEs, changelogs), which presents an indirect prompt injection surface. However, the risk is mitigated as these operations are performed on the user's local repository.
Ingestion points: scripts/check-manifests.sh reads package.json, pyproject.toml, Cargo.toml, and other manifest files.
Boundary markers: Absent.
Capability inventory: Shell script execution, file system modifications via scripts/bump-manifests.sh, and GitHub API access via gh CLI.
Sanitization: Absent; the skill relies on the integrity of the local repository environment.
[SAFE]: The skill uses an embedded Python snippet within shell scripts to handle complex data formats like TOML. This is implemented as a static HEREDOC and does not execute untrusted external code.
Evidence: scripts/bump-manifests.sh and scripts/check-manifests.sh use `python3
<<'PY'` for safe, local parsing.

git-guard