The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructions direct the agent to generate and execute its own TypeScript code locally to filter and process search results, rather than relying on static tool definitions.
Evidence: SKILL.md documentation for "Code Mode Pattern" and "Option 2: Inline TypeScript" examples which instruct the agent to "Write TypeScript code... Execute the code via Bash".
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted code snippets retrieved from millions of public GitHub repositories, which could contain instructions aimed at manipulating the agent's behavior when it reads or analyzes the search results.
Ingestion points: scripts/search.ts and SKILL.md examples fetch data from the grep.app API.
Boundary markers: The skill does not define specific boundaries or instructions to distinguish untrusted code snippets from legitimate commands or instructions.
Capability inventory: The skill utilizes shell access for bun, curl, and jq across its scripts and examples.
Sanitization: scripts/search.ts performs basic HTML tag stripping and entity replacement but does not sanitize the content for potential prompt injection attacks.
[EXTERNAL_DOWNLOADS]: The skill fetches search results and repository metadata from the external grep.app service.
Evidence: scripts/search.ts and SKILL.md examples make network calls to https://grep.app/api/search.
[COMMAND_EXECUTION]: The skill uses local shell commands to execute search scripts and manipulate JSON data retrieved from the web.
Evidence: Implementation examples in SKILL.md use bun run, curl, and jq.

github-code-search