The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon data from external, potentially untrusted sources such as GitHub issue bodies, issue comments, and PR review comments.
Ingestion points: Instructions in references/autonomous-mode.md describe reading issue content to parse requirements, check status, and identify 'Pre-existing Changes'.
Boundary markers: The skill includes a 'Security Note' and 'Security Warning' advising the agent to treat issue content as untrusted and to ignore instructions that contradict the skill's defined workflow.
Capability inventory: The skill has extensive capabilities including executing shell commands, performing git operations (worktree management, applying diffs, committing, pushing), and interacting with GitHub/GitLab APIs via CLI wrappers.
Sanitization: The skill provides logic to skip applying pre-existing changes if the issue author is not a trusted repository collaborator.
[COMMAND_EXECUTION]: The workflow relies on executing various system and development commands.
Evidence: SKILL.md and references/autonomous-mode.md contain instructions for git worktree, git apply, git cherry-pick, and running project-specific test suites (e.g., npm run test) inside worktrees.
Evidence: Use of scripts/gh-as-user.sh to post comments and trigger bots as a specific identity.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of project dependencies during worktree setup.
Evidence: Step 2 in SKILL.md includes npm install (or equivalent package manager commands) after creating a new worktree.

autonomous-dev