The Agent Skills Directory

Command Execution: The skill uses shell commands like cat, gh api, and jq to read configurations and fetch repository data. These operations are restricted to read-only access to the CI environment as per the skill's stated rules.
Dynamic Execution: The instructions include a pre-flight verification step that executes a Python module from a local directory (.apache-magpie-local). This allows for dynamic checks of the project environment before the triage process begins.
Indirect Prompt Injection Surface: The skill processes external data retrieved from GitHub, such as workflow names and commit messages. There is a potential for this data to contain malicious directives. The skill addresses this by including explicit "Golden Rules" that instruct the agent to treat all fetched content strictly as data and ignore any embedded commands.
Ingestion points: Data enters the agent context through gh api calls in the Data Collection section of SKILL.md.
Boundary markers: Explicit instructions in the "Golden Rules" section define the boundary between system instructions and external data.
Capability inventory: The skill has capabilities for file reading (cat) and external API interaction (gh api).
Sanitization: External data is processed using jq to parse JSON, helping ensure data is handled structurally rather than as raw text.

flaky-test-triage