The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the repository being analyzed (such as README files, package manifests, and source code) and interpolates it into the generated CLAUDE.md file.
Ingestion points: Phase 1 (SCAN) reads various project files including README.md, go.mod, package.json, and Makefile to gather facts.
Boundary markers: The skill uses markdown templates and explicit phase-based instructions, but it lacks specific safety warnings to the agent about ignoring potential prompt injection instructions embedded within the analyzed repository files.
Capability inventory: The skill has Write access to create files and Bash/Grep access to execute local shell commands for inspection.
Sanitization: The skill implements a 'Banned Generic Phrases' check and placeholder detection in Phase 4 (VALIDATE), which helps ensure quality but does not specifically filter for malicious prompt injection patterns in the source data.
[COMMAND_EXECUTION]: The skill uses the Bash tool to perform repository inspection. The commands used (ls, grep, cat, head, test, which, diff) are restricted to local read-only analysis and comparison, representing a low risk consistent with the skill's primary purpose of repository mapping.

generate-claudemd