The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's workflow involves sub-agents analyzing Claude Code session transcripts which contain untrusted user and agent content. This ingestion of external data creates a surface for indirect prompt injection, where malicious instructions embedded in transcripts could theoretically influence the sub-agents' analysis or recommendations. Ingestion points include the transcripts.json file processed by the impact, always-on, portfolio, and planner analysts. There are no explicit boundary markers or instructions to ignore embedded commands within the transcript data. The skill possesses the capability to modify markdown files through scripts/apply_recommendations.py and execute shell commands via the coordinator workflow. No sanitization or filtering of the transcript content is performed before it is provided to the agent context. However, the risk is mitigated by the fact that all recommendations must be manually reviewed and confirmed by the user before being applied.
[COMMAND_EXECUTION]: The skill executes several local commands and scripts as part of its auditing process. The SKILL.md workflow uses shell commands for workspace management (mkdir, date) and to launch scripts. Additionally, scripts/collect_docs.py utilizes subprocess.run to call git log and git rev-list to gather documentation freshness data. These operations are scoped to the local environment and the target project directory.

docs-auditor