The Agent Skills Directory

[COMMAND_EXECUTION]: The skill automates several CLI-based operations using standard engineering tools.
It executes git commit and git push to persist implementation changes and review fixes to remote repositories.
It uses gh pr view and gh pr edit to autonomously manage pull request descriptions based on analysis results.
It dynamically resolves Git remotes and sets upstream tracking branches without user intervention.
[DATA_EXFILTRATION]: The skill transmits local data and code artifacts to external platforms as part of its handoff process.
It pushes local review findings to remote Git branches in the docs/ directory.
It sends findings to external project management services (GitHub Issues, Linear, Jira) via CLI tools, MCP interfaces, or APIs.
These operations are performed in 'Non-interactive mode,' which explicitly skips user confirmation before transmitting data.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by processing and re-broadcasting untrusted data from the local environment.
Ingestion points: The skill ingests user-provided $ARGUMENTS and the output (artifacts/summaries) from the ce-code-review and ce-plan skills.
Boundary markers: There are no explicit delimiters or instruction-isolation markers used when interpolating these findings into PR bodies or tracker tickets.
Capability inventory: The skill has the capability to write to the local filesystem, commit to Git, push to remote servers, and edit external pull requests.
Sanitization: The instructions do not specify any validation or sanitization logic for the content extracted from review findings before it is published to external platforms.

lfg