The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external PR reviews (via the sanctum:pr-review dependency) and store it as project knowledge. This content, which originates from PR authors and reviewers, could contain hidden instructions intended to influence the agent when the data is later retrieved and presented.
Ingestion points: Findings and discussion context are extracted from PR reviews and stored in files under the review-chamber/ directory.
Boundary markers: The skill uses structured YAML frontmatter and specific Markdown headings (e.g., ### Decision, ### Context) to organize data. While this structure helps the agent distinguish between different types of information, it does not explicitly sanitize the content against adversarial instructions.
Capability inventory: The skill possesses the ability to write to the filesystem (to store entries) and execute local scripts (such as scripts/palace_manager.py) to manage the knowledge base.
Sanitization: The provided documentation and module logic do not detail any specific sanitization or escaping of the PR content before it is interpolated into palace entries.
[COMMAND_EXECUTION]: The skill's workflow and CLI reference include the execution of local scripts for data management.
Evidence: The skill frequently references invoking python scripts/palace_manager.py for searching, listing, and exporting review findings. These commands are used for administrative tasks within the skill's defined scope.

review-chamber