The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to research external web sources and code repositories, which creates a surface for indirect prompt injection. The skill mitigates this risk by enforcing structured evidence capture and providing a validation checklist for the agent to verify findings against primary sources.
Ingestion points: Ingests content from the web and remote repositories in Step 3: Research + Evidence Capture.
Boundary markers: Uses standardized evidence file templates and markdown code blocks to delimit external content.
Capability inventory: Local script execution via bun (scripts/generate-catalogue.ts), file system modification (scripts/normalize-reports.ts), and agent-controlled web search.
Sanitization: Synthesis steps require the orchestrator to resolve conflicts and verify evidence before writing final reports.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill executes local TypeScript scripts using the bun runtime and depends on the standard gray-matter package for parsing metadata. It also facilitates cloning of official or specified repositories for codebase research.
Evidence: Scripts are located in the scripts/ directory and managed via package.json. Git operations are documented in references/source-code-research.md.
[COMMAND_EXECUTION]: The skill uses shell commands through execSync in its internal maintenance scripts to perform git operations and file searching.
Evidence: scripts/normalize-reports.ts uses grep for cross-reference updates and git rev-parse to identify the repository root.

research