The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from a local codebase, which is an untrusted source. Maliciously crafted code or comments could attempt to influence the agent's behavior during the search and synthesis process.
Ingestion points: Files within the local project directory processed by search tools such as grep and codebase_search.
Boundary markers: Not explicitly defined; however, the skill provides a critical security instruction for the agent to verify cited locations manually before relaying load-bearing conclusions.
Capability inventory: Routing to specific search plugins and executing git commands via bash for history/blame.
Sanitization: No automated sanitization is described; the skill relies on the agent's manual verification of citations to mitigate the risk of misinterpretation or malicious synthesis.

code-search