The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for the agent to perform development tasks by executing shell commands and local CLI tools.
Evidence: The skill utilizes git, gh (GitHub CLI), jq, and a custom wspace utility to manage repositories and automate the 'Goal loop' and 'Pipeline' workflows defined in SKILL.md and references/COMMANDS.md.
[SAFE]: The skill implements a surface for indirect prompt injection by processing external GitHub issue data, but employs structured data handling to mitigate risk.
Ingestion points: The SCAN phase in SKILL.md fetches issue metadata (number, title, repository) from GitHub using gh search issues.
Boundary markers: The process uses the --json flag to ensure data is returned in a structured format rather than raw text.
Capability inventory: The agent performs local file system operations, Git management, and PR creation based on the ingested data.
Sanitization: The use of jq to extract specific fields (number, repository) ensures that only expected data influences downstream command execution, limiting the impact of malicious content in issue titles.
[SAFE]: The skill handles sensitive environment variables using local-only synchronization practices.
Evidence: The wspace env sync command manages project secrets by copying them from a central, gitignored secrets/ directory to specific worktrees. This approach avoids hardcoded credentials and prevents secrets from being committed to version control, which is an established safe practice.

wspace