The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute arbitrary shell commands defined in the .candid/config.json configuration file, specifically for installCommand, buildCommand, testCommand, and postMergeCommand. This enables the agent to perform extensive operations on the host system.
[REMOTE_CODE_EXECUTION]: The execution of user-defined commands, such as the curl example provided for the postMergeCommand, allows for downloading and potentially executing remote scripts or triggering external webhooks during the workflow.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the repository environment, including branch names and commit subjects, and user-provided context via additionalPrompt. These inputs are interpolated into pull request descriptions and passed to sub-tasks like code reviews without explicit sanitization.
Ingestion points: Reads additionalPrompt from configuration; extracts currentBranch and commitSubjects from git logs and branch metadata.
Boundary markers: Lacks explicit delimiters or instructions for the agent to ignore potentially malicious commands embedded in the interpolated content.
Capability inventory: The skill has access to shell execution, GitHub CLI (gh), git, and the Linear MCP tool.
Sanitization: No evidence of input validation or escaping for the additionalPrompt before it is passed to the candid-loop tool.
[EXTERNAL_DOWNLOADS]: The workflow encourages and executes the installation of external dependencies using package managers like npm or pnpm, which involve fetching code from public registries at runtime.

candid-ship