The Agent Skills Directory

[SAFE]: No malicious patterns such as obfuscation, credential exfiltration, or unauthorized network operations were detected in the skill files.
[COMMAND_EXECUTION]: The validation script scripts/validate.sh executes a local helper script (heal.sh) and performs string checks using grep. The execution uses a relative path within the repository structure, which is standard for build/validation tasks and does not involve untrusted data.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data such as caller-outcome and proposed-process-work. However, it includes strong boundary instructions, requiring the agent to return a specific YAML shape and explicitly prohibiting the creation of files or tracker entries, which effectively mitigates exploitation risks.

anti-ceremony