The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to execute shell commands for environment analysis and file discovery. These commands include user-provided parameters such as file paths and git commit hashes.
Evidence: SKILL.md (Phase 0.5) executes git status --porcelain -- <resolved paths> and reference/flows/mixed.md executes git diff --name-only <hash>^..<hash> and a complex find command.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from external files (e.g., source code, documentation, and chat logs), which presents an attack surface where embedded instructions could attempt to influence the agent's behavior during the 'humanization' process.
Evidence Chain:
Ingestion points: The skill uses the Read tool to load content from user-specified files or git commits.
Boundary markers: The prompts for sub-agents (defined in reference/flows/mixed.md) do not use robust delimiters to isolate file content from the task instructions.
Capability inventory: The skill and its sub-agents have access to sensitive tools including Write, Edit, Bash, and Agent (via Task).
Sanitization: The skill includes a 'PASS 1 STRIP' phase (documented in reference/ai-patterns.md) designed to remove AI self-references and prompt residue from the input text.
[DYNAMIC_EXECUTION]: The skill dynamically generates tasks for sub-agents using the Agent tool. These tasks are constructed by interpolating system templates with user-provided custom instructions and environment variables.
Evidence: In reference/flows/mixed.md, the Task prompt is built by concatenating raw [CUSTOM_INSTRUCTIONS_IF_ANY] and expanding the ${CLAUDE_PLUGIN_ROOT} variable.

text-human