The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user prompts regarding project goals and from existing codebase files during brownfield analysis. This data is incorporated into generated documentation like product.md without explicit sanitization.
Ingestion points: SKILL.md (Project Goal prompt in Section 2 and codebase architecture scan for brownfield projects).
Boundary markers: No specific delimiters or safety instructions are defined to isolate processed external content from the agent's instructions.
Capability inventory: The skill possesses the ability to execute shell commands (git, curl, mkdir, python3) and perform filesystem write operations.
Sanitization: No instructions are provided for filtering, escaping, or validating the external inputs before interpolation into prompt contexts.
[EXTERNAL_DOWNLOADS]: The skill retrieves agent skill files from remote repositories using curl as part of its automated setup process.
Evidence: Section 2.6 of SKILL.md outlines a workflow for downloading SKILL.md files to the local .agents/skills/ directory.
Sources: The downloads target the official Firebase GitHub organization and the author's own extension repositories, both of which are recognized as reputable sources.
[COMMAND_EXECUTION]: The skill utilizes local system utilities to perform project initialization and state management.
Evidence: SKILL.md instructs the agent to run python3 scripts/resume.py to check state and various git commands (init, status, ls-files, add, commit, notes) to manage project scaffolding.

conductor-setup