The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute shell commands for git operations, interact with forge CLIs (gh, glab), and run repository-specific verification commands defined in the project's configuration file.
[DATA_EXFILTRATION]: The skill is designed to push local code changes and issue updates to remote platforms such as GitHub, GitLab, and Linear. While this is intended behavior, it involves the transmission of potentially sensitive repository data to external services.
[INDIRECT_PROMPT_INJECTION]: The skill processes requirements and feedback from issue tracker content which can be modified by untrusted third parties, creating a surface for indirect prompt injection attacks.
Ingestion points: The agent reads issue bodies and comments via forge APIs or Linear MCP to determine work scope (SKILL.md Step 6).
Boundary markers: Instructions do not provide specific delimiters or ignore-instructions warnings to encapsulate ingested text.
Capability inventory: The agent has extensive permissions including shell access, file editing, and the ability to push commits to remote branches.
Sanitization: No validation or sanitization of external issue content is performed prior to execution.
[EXTERNAL_DOWNLOADS]: When operating in parallel mode, the skill executes dependency installation commands (e.g., npm ci, pnpm install, composer install) targeting well-known official registries based on the project's committed lockfiles.

work-implement