The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from issue trackers and repository content, creating a potential surface for indirect prompt injection attacks where malicious instructions could be embedded in bug reports or code comments.
Ingestion points: The skill reads GitHub/tracker issue titles, bodies, and comments, as well as repository source code, documentation, and CI logs (documented in SKILL.md and references/agentic-setup.md).
Boundary markers: The skill defines an explicit 'Untrusted content boundary' in references/agentic-setup.md that instructs the agent to treat external content as data rather than instructions and provides specific examples of directives to ignore and report (e.g., 'ignore previous instructions').
Capability inventory: The skill performs git operations, modifies source files, creates pull requests, and executes project validation/test suites through delegated chain skills (e.g., om-fix, om-auto-review-pr).
Sanitization: Implements validation of externally-sourced values (numeric checks for IDs, regex-based slug generation) before shell or path interpolation, as specified in references/worktree-setup.md and references/agentic-setup.md.
[COMMAND_EXECUTION]: The skill performs shell commands to manage isolated git worktrees and install project dependencies as part of the fixing process.
Evidence: references/worktree-setup.md contains shell script templates for git worktree add and git checkout operations. The skill explicitly mandates strict sanitization of variables like {issueId} and {slug} to prevent command injection from tracker data.
[EXTERNAL_DOWNLOADS]: The skill interacts with external issue trackers and repository hosting services to fetch work items and submit changes, which involves connecting to remote APIs.
Evidence: SKILL.md and references/claim-pr.md describe operations such as get-issue, search-prs, and create-pr via a tracker descriptor. These operations are restricted to the repository's configured environment and follow the operator-vouched configuration defined in the agentic setup.

om-auto-fix-issue