The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs local shell command execution using tools like git, the GitHub CLI (gh), and the Datadog pup utility. These commands are used to resolve repository metadata, identify branch names, and trigger CI workflow retries as part of its core triage functionality.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data, such as branch names and repository identifiers, and interpolates them into shell commands and Datadog event queries.
Ingestion points: Untrusted data enters the agent context from git remote get-url origin and user-supplied parameters in SKILL.md (PR branch, PR number, Repository).
Boundary markers: The skill lacks explicit delimiters or warnings to instruct the agent to ignore potentially malicious instructions embedded within the branch names or repository metadata.
Capability inventory: The skill possesses capabilities for command execution (via git, pup, gh) and network-based data retrieval through Datadog tools across all its workflow steps.
Sanitization: The instructions explicitly require the agent to normalize and lowercase repository IDs and branch names, providing a baseline level of sanitization for the ingested data.

unblock-pr