The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands to detect the repository's configuration and history.
Evidence: Uses git rev-parse --show-toplevel to identify the repository root and git log --oneline -30 to sample recent commit messages for pattern detection.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local repository files which could theoretically contain malicious instructions, though this is a standard risk for development tools.
Ingestion points: Reads package.json (specifically scripts and workspaces), tsconfig.json, lockfiles, and the output of git log from SKILL.md.
Boundary markers: Absent; the skill does not use specific delimiters to isolate data ingested from these files.
Capability inventory: The skill uses the Write tool to create a new skill file at .claude/skills/do-work/SKILL.md. The resulting skill is designed to execute shell commands (e.g., test and lint scripts).
Sanitization: Absent; the skill directly interpolates values extracted from package.json into the generated skill's workflow templates.
[DYNAMIC_EXECUTION]: The skill generates a new workflow skill by populating a template with configuration data extracted at runtime.
Evidence: The skill resolves placeholders such as {{FEEDBACK_COMMANDS}} in templates/do-work.template.md by extracting script names from the project's package.json and writes the final markdown to a new skill file.

generate-do-work