The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes common developer tools such as git and gh (GitHub CLI) to retrieve repository information and manage branches. These operations are limited to standard version control tasks.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes data that could potentially be influenced by external actors via Git metadata.
Ingestion points: Ingests repository state and history through git log, git diff, and gh pr view as defined in SKILL.md.
Boundary markers: While explicit delimiters are not defined for tool outputs, the skill's multi-gate procedure requires the agent to interpret data as evidence for a specific decision matrix rather than direct instructions.
Capability inventory: Capabilities are restricted to local and remote Git branch management (e.g., deletion and switching); no arbitrary code execution or generalized network access is requested.
Sanitization: The skill incorporates mandatory 'Handoff' and 'Safety Check' gates (Gate 5 and Gate 6-3) which require human review and confirmation before any staging, committing, or branch deletion occurs.
[SAFE]: No evidence of credential harvesting, unauthorized data exfiltration, or obfuscated code was detected. The skill follows security best practices by recommending specific file staging and avoiding broad commands like git add ..

branch-context-check