The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill explicitly instructs agents to read environment configuration files and search for API/SDK imports (e.g., Stripe, Supabase, AWS) to identify external integrations. Accessing environment files and documenting secret storage locations exposes sensitive configuration data.
Evidence: Instructions in SKILL.md to run ls -la .env* and search for credential-related imports.
Evidence: INTEGRATIONS.md.template requires identifying environment variable names and secret storage locations.
[INDIRECT_PROMPT_INJECTION]: The skill processes the entire codebase as untrusted input to generate its reports. Maliciously crafted content within the codebase, such as instructions embedded in comments or documentation, could influence the behavior of the mapping agents.
Ingestion points: The entire codebase is explored via cat, rg, find, and fd commands defined in SKILL.md.
Boundary markers: The prompts for the parallel agents (Tech, Architecture, Quality, Concerns) lack delimiters or instructions to ignore embedded directives in the files they analyze.
Capability inventory: The skill can create directories, write markdown files, execute shell utilities, and spawn child agents using the Task tool.
Sanitization: There is no evidence of sanitization or filtering of the codebase content before it is processed by the agents.
[COMMAND_EXECUTION]: The skill makes extensive use of shell commands (mkdir, ls, wc, fd, find, rg, grep, git) for codebase exploration and management. Executing these tools on an untrusted codebase carries inherent risks, especially when committing changes or listing large directory structures.

codemap