The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a local Node.js script (scripts/resolve-feature.js) to identify feature context. The instructions include a specific security directive to pass the feature key as a separate token and avoid shell interpolation, which effectively prevents command injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an inherent attack surface for indirect prompt injection due to its core function of reading and processing external codebase content.
Ingestion points: Data is ingested from files categorized as current_authority, design_records, and work_records via the feature resolver.
Boundary markers: The skill does not use explicit delimiters for processed content but implements a provenance manifest that records the source file paths and SHAs to maintain a record of document lineage.
Capability inventory: The skill's operational capabilities are restricted to Read, Write, Edit, and filtered Bash commands (node and git).
Sanitization: The skill follows strict "Redaction Rules" to identify and replace sensitive data—such as API keys, secrets, internal endpoints, and database connection strings—with symbolic placeholders (e.g., ${ENV_VAR_NAME}) before output is written to disk.

runbook