The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external project data including tool files, test suites, and manifest files (.discover.json) to generate reports. This creates a surface for indirect prompt injection if project content is maliciously crafted.
Ingestion points: Reads directory structures and file contents from TOOLS_DIR, EVALS_DIR, API_GENERATED_DIR, and the project root manifest in scripts/count-tools.ts.
Boundary markers: None identified; output is generated without explicit delimiters or warnings to the agent regarding embedded content.
Capability inventory: Executes local scripts via npx tsx, reads arbitrary files in the project structure, and writes markdown reports to specified paths.
Sanitization: Extracted strings (such as tool names and collection names) are not sanitized before being printed to the console or written to files.
[COMMAND_EXECUTION]: The skill uses Bash to execute a TypeScript counting script. While the script is part of the skill, it allows the agent to define environment variables like OUTPUT_FILE and PROJECT_ROOT, which control where the script reads from and writes to on the file system.

count-mcp-tools