The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted BRD content and audit reports to automatically generate and apply patches.
Ingestion points: Reads .aidoc/audit/01_BRD-audit.md and target BRD files from the local filesystem to identify and fix issues.
Boundary markers: The skill instructions do not define clear delimiters or instructions to ignore embedded prompts within the documents it remediates, which could allow a malicious document to influence agent behavior.
Capability inventory: The skill has capabilities to write files, create directories, and execute shell commands via subagents or direct invocation.
Sanitization: There is no explicit sanitization step for the content of the BRD sections before they are used as inputs for patches or validation tasks.
[COMMAND_EXECUTION]: The skill invokes an external Python module using document-derived strings as command-line arguments.
Evidence: The skill calls python -m sdd_doc_lint.rehash using parameters such as --title T and --description D which are populated from the BRD content being audited.
Mechanism: Shell-based execution of local utilities which, if not handled carefully by the underlying script, could be susceptible to argument injection if the document content contains shell-sensitive characters.

doc-brd-fixer