The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill audits run logs which contain data generated during previous AI sessions (e.g., markdown reports, JSON metadata, and OOS blocks). This data is potentially influenced by external PR content or previous agent outputs. The skill parses these logs and includes the findings in new GitHub issues and comments.
Ingestion points: The scan-run subcommand reads from multiple files including voting-tally.md, review-findings-full.jsonl, round-meta.json, and execution-issues.ndjson located in the run-log directories.
Boundary markers: No explicit delimiters or warnings to ignore embedded instructions within the logs were identified in the instructions.
Capability inventory: The skill uses the Bash tool to execute larch.sh for filing issues (issue create-one), posting comments (audit-runs comment), and closing issues (audit-runs close-priors).
Sanitization: No explicit sanitization, escaping, or instruction-aware filtering of the ingested log content is described before it is processed or interpolated into issue bodies.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to run a centralized utility script (larch.sh) located in the plugin's scripts directory. This script performs all core logic, including git operations, log processing, and GitHub API interactions via subcommands such as preflight, scan-run, and close-priors. These operations are consistent with the skill's stated purpose of auditing repository activity.

audit-runs