The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from external sources, specifically GitHub issue reports and local markdown files.
Ingestion points: The ARGUMENT variable in SKILL.md (Step 1) allows the agent to fetch content using gh issue view or direct file reads.
Boundary markers: The skill attempts to implement a logical boundary by using a separate general-purpose subagent to extract facts and explicitly forbidding the main agent from reading the raw source. However, there are no structural delimiters or escaping mechanisms to prevent a malicious report from influencing the subagent's output.
Capability inventory: The skill can write to the filesystem, execute shell commands (tests and pre-commit), and spawn additional agents.
Sanitization: There is a list of exclusions (suspected causes, PR links, etc.) provided to the subagent to act as a manual filter, though this does not programmatically sanitize the input against sophisticated injections hidden in verbatim "Reproduction" or "Evidence" blocks.
[COMMAND_EXECUTION]: The skill uses several command-line tools to perform its tasks.
Evidence: Execution of gh issue view to retrieve external data, git diff to view changes, and pre-commit run --files <changed files> which executes project-specific hooks. These hooks typically run arbitrary code (linters, formatters) defined within the repository's environment.

fix-issue