The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and processes content from GitHub issues (specifically from agent-sh/agnix). Maliciously crafted issue bodies could attempt to influence the agent's behavior during the implementation phase.
Ingestion points: The skill uses larch.sh gh agnix-issue to retrieve issue titles and bodies from GitHub.
Boundary markers: Step 4 of SKILL.md notes that the implement skill wraps the issue title, body, and plan in <reviewer_issue_title>, <reviewer_issue_body>, and <reviewer_plan> delimiters to isolate untrusted content.
Capability inventory: The skill utilizes Bash for environment validation and label provisioning, and the Skill tool to trigger the implement skill.
Sanitization: The input <upstream-issue-number> is validated as an integer using a regular expression, and external issue data is parsed using jq.
[COMMAND_EXECUTION]: The skill performs several local command executions using scripts/larch.sh. These commands are used to verify the git remote, fetch issue details from GitHub, and manage repository labels. These operations are restricted to the context of the agent-sh/agnix repository.

agnix-fix