The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from .candid/last-review.json, which is output from the candid-review skill. This data includes issue titles and IDs derived from the analyzed source code.
Ingestion points: The skill reads the review state from .candid/last-review.json in Step 3.2.
Boundary markers: No explicit delimiters or boundary markers are mentioned for processing the ingested JSON content.
Capability inventory: The skill possesses file system access (read/write via cat, jq, and mv) and the ability to invoke other skills (/candid-review).
Sanitization: The skill parses the JSON structure but does not specify sanitization for issue IDs or titles before they are used in Step 3.7.
[COMMAND_EXECUTION]: The skill uses jq and shell commands (mv) to manage configuration files. These operations are restricted to the local project directory and the user's home directory configuration (~/.candid/config.json), which is consistent with its role as a developer automation tool.
[SAFE]: The operations performed by the skill, including local configuration management and tool iteration, are aligned with its stated purpose. No high-risk command execution, remote code execution, or credential exfiltration attempts were identified.

candid-loop