The Agent Skills Directory

[SAFE]: The skill implements secret isolation by providing specific instructions and examples for managing credentials via environment variables and excluding sensitive files like .env and .token-cache from version control using .gitignore.
[SAFE]: The token caching implementation in src/token-cache.ts uses restrictive file system permissions (0o600), ensuring that access to cached OAuth tokens is limited to the local file owner.
[SAFE]: A redaction pattern is included in src/request-logger.ts to prevent the accidental logging of client_secret values in plain text within URLs.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a data ingestion surface by capturing and logging external API response data.
Ingestion points: API response bodies are recorded in logs/navan-api.log via the request logger and in tests/fixtures/ via the fixture recording script.
Boundary markers: There are no explicit delimiters or specific instructions for downstream processes to ignore instructions that might be embedded in the logged API data.
Capability inventory: The skill utilizes Read, Write, Edit, Grep, and Bash tools to configure the project structure and execute scripts.
Sanitization: While the code redacts secrets from the URL, it stores a preview of the response body and full fixture data without additional filtering or sanitization of potential injection vectors in the data content.

navan-local-dev-loop