The Agent Skills Directory

[SAFE]: The skill includes explicit instructions in commit.md and work.md to exclude sensitive files such as .env, credentials, and private keys from git staging and commits, protecting the user's secrets.
[EXTERNAL_DOWNLOADS]: The version.md action performs a network request to the vendor's official GitHub repository to compare local and remote versions. This is a standard update-checking mechanism using a well-known service.
[COMMAND_EXECUTION]: The orchestrator manages the task queue and archive using standard filesystem and git commands. These operations are scoped to the project's do-work/ directory and follow strictly defined patterns.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes verbatim user input that is later used as context for implementation agents. * Ingestion points: User requests are captured verbatim in do-work/user-requests/UR-NNN/input.md and do-work/REQ-*.md files. * Boundary markers: No specific boundary markers or 'ignore' instructions are used to delimit user-provided content from agent instructions. * Capability inventory: The orchestrator spawns subagents with shell execution and file modification capabilities to fulfill the requests. * Sanitization: No sanitization or filtering is applied to the captured verbatim input.

do-work