The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests test failure output (stdout/stderr) to parse failure details and diagnose root causes. This data is untrusted as it originates from external test runs which could be manipulated to include malicious instructions. \n
Ingestion points: Failure output is processed via the failure-parser regex in SKILL.md (Phase 1). \n
Boundary markers: The skill does not use explicit delimiters or boundary markers to separate the untrusted test output from the agent's internal instruction context. \n
Capability inventory: The skill possesses shell execution capabilities (to run tests) and file modification capabilities (via the Edit tool). \n
Sanitization: While the skill uses regex to extract specific fields, it does not perform sanitization or validation of the extracted test names or messages before using them in reasoning or command construction.
[COMMAND_EXECUTION]: The skill performs multiple shell operations to identify projects and run tests. \n
Evidence: SKILL.md and rules/project-keying.md utilize commands such as git remote get-url, git rev-parse --show-toplevel, and readlink to establish project identity. \n
Evidence: Phase 4 of SKILL.md executes a single-test-command derived from configuration templates.
[DYNAMIC_EXECUTION]: The skill constructs and executes shell commands at runtime by interpolating variables (such as {file} and {name}) into templates stored in surface configuration files. \n
Evidence: The orchestration logic in SKILL.md (Phase 4) and the stack detectors (e.g., detectors/vitest.md) define command templates that are populated with test names and paths parsed from previous runs. This creates a risk of command injection if a test name contains shell metacharacters.

test-auto-fix