The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses rg (ripgrep) to search for existing diagnosis files and a local utility zg index to refresh a semantic search index. Both operations are constrained to the local ~/ops directory.
[INDIRECT_PROMPT_INJECTION]: The skill ingests raw data from bug diagnosis runs, including command outputs, logs, and existing markdown files, which could serve as a vector for indirect prompt injection if the source data is untrusted.
Ingestion points: Bug diagnosis reports, logs, and existing markdown files in ~/ops/*.md.
Boundary markers: No explicit delimiters are used to wrap ingested log data or command outputs to distinguish them from instructions.
Capability inventory: File system write access, execution of local search/indexing tools, and git commit operations.
Sanitization: The instructions require evidence to be pasted "as-is", which may include raw, unsanitized strings from logs or external system outputs.

diag-write