The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from local codebase files to generate technical documentation and configuration files.
Ingestion points: Codebase files are read using tools like grep, find, and cat in SKILL.md (Steps 3, 5, 8) and by parallel sub-agents defined in reference/analysis-prompts.md.
Boundary markers: The instructions for sub-agents and the main workflow do not include explicit delimiters or instructions to ignore potential commands embedded within the analyzed source code.
Capability inventory: The skill uses shell execution for analysis (Step 5), creates new files (Step 11.1), and can invoke other agent skills (Step 11.2).
Sanitization: There is no evidence of filtering or sanitizing extracted code snippets or patterns before they are synthesized into the final Technical.md and config.json files.
[COMMAND_EXECUTION]: The skill makes extensive use of local shell commands to audit project structure and configuration.
Evidence: Uses ls, cat, find, grep, sed, awk, git, and jq throughout the workflow to audit the filesystem and configuration files.
Evidence: The file reference/analysis-commands.md lists various bash recipes used for architecture detection and pattern analysis.

candid-init