The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from GitHub PR comments and descriptions to generate new issues, creating a vulnerability to indirect prompt injection.
Ingestion points: SKILL.md Step 2 (fetching comments via get-pr-comment and list-issue-comments) and Step 5 (fetching PR bodies via get-pr).
Boundary markers: references/agentic-setup.md includes explicit instructions for the agent to treat repository and tracker content as data rather than instructions and to report suspected injection attempts.
Capability inventory: The skill has the capability to create new issues and post comments using create-issue and comment-pr tools.
Sanitization: SKILL.md Step 2 requires the redaction of credential-like strings (tokens, API keys, etc.) from external text before it is quoted in the generated issue.
[DYNAMIC_EXECUTION]: The skill is configured to load and apply supplemental instructions from the repository being analyzed.
Evidence: references/agentic-setup.md Step 3 instructs the agent to apply .ai/skills/om-followup-issue-from-pr/SKILL.md from the local repository as an extension to the current skill instructions.
[COMMAND_EXECUTION]: The skill uses local shell commands to handle configuration data.
Evidence: references/agentic-setup.md contains shell snippets using jq to parse the .ai/agentic.config.json file for settings like baseBranch and labels.enabled.

om-followup-issue-from-pr