The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it parses content from GitHub pull request comments and Linear issues to generate its reports and action plans. A malicious actor could influence the agent's behavior or recommendations by embedding deceptive data or instructions within these external sources (e.g., mimicking a Greptile review comment or faking a confidence score to trigger an auto-merge recommendation).
Ingestion points: GitHub issue comments (gh api .../comments), PR review states, and Linear issue details (mcp__linear-server__list_issues).
Boundary markers: None present; the skill does not instruct the agent to ignore or delimit instructions found within the ingested data.
Capability inventory: GitHub PR merging (gh pr merge), git push operations (git push --force-with-lease), and automated Slack messaging.
Sanitization: None present; content is processed and presented directly in reports.
[DATA_EXFILTRATION]: The skill automatically transmits project status data, including repository names, pull request titles, and Linear issue identifiers, to an external Slack channel via the Slack MCP tool. While this is a functional requirement, the automatic execution without a specific user confirmation per-message reduces oversight of what project metadata is sent to the remote service.
[COMMAND_EXECUTION]: The skill relies on executing shell commands (gh, git, cat, jq) that utilize dynamic variables derived from the local environment and external repository state. While these are necessary for project management, the reliance on shell execution for critical actions like merging and rebasing represents a high-capability surface that depends on the integrity of the data being processed.

review-open-prs