The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The 'Requirement Drift Detection' procedure in SKILL.md explicitly instructs the agent to read and follow instructions found in issue comments that may have been posted after the initial implementation.
Ingestion points: Fetches issue comments using gh issue view --json comments.
Boundary markers: The instructions do not define specific delimiters or security warnings to treat the comment content as untrusted data.
Capability inventory: The skill has the ability to perform git push --force-with-lease, execute shell scripts, and post PR/issue comments.
Sanitization: No sanitization or validation of the comment content is mentioned before the agent uses it to influence its review decision.
[COMMAND_EXECUTION]: The skill is heavily reliant on executing shell commands and scripts to perform its tasks.
Executes project-specific scripts including scripts/gh-as-user.sh, scripts/mark-issue-checkbox.sh, scripts/post-verdict.sh, and scripts/upload-screenshot.sh.
Performs repository modifications using git worktree, git rebase, and git push --force-with-lease as described in references/merge-conflict-resolution.md.
Supports a command-mode E2E verification (references/e2e-command-mode.md) that executes an arbitrary, project-supplied string assigned to E2E_COMMAND.
[CREDENTIALS_UNSAFE]: The skill documentation and scripts reference the use of sensitive authentication tokens and credentials.
scripts/upload-screenshot.sh requires GH_TOKEN or GITLAB_TOKEN to be present in the environment for code-host interactions.
references/e2e-verification.md specifies that E2E_TEST_USER_EMAIL and E2E_TEST_USER_PASSWORD environment variables are used for browser-based login automation.

autonomous-review