The Agent Skills Directory

[SAFE]: The skill acts as a redirection layer for legacy support, delegating core logic to a separate, canonical planning skill. It contains no direct malicious code or deceptive instructions.
[COMMAND_EXECUTION]: The skill includes instructions to execute a repository-local script (utilities/skill-builder/scripts/record_skill_feedback.py) to log interaction metadata. This is part of a standard quality assurance and feedback loop within the developer environment.
[DATA_EXFILTRATION]: A feedback protocol is defined to capture decision metrics (e.g., acceptance, outcome, confidence). This data is stored using a local script and is used for internal process analytics rather than unauthorized external transmission.
[PROMPT_INJECTION]: As a planning tool, the skill is designed to ingest and process user-provided technical requirements and specifications.
Ingestion points: User-supplied feature requirements and repository context enter the workflow via input placeholders defined in the contract and main skill file.
Boundary markers: The skill does not explicitly define unique delimiters for untrusted input, but it operates as a wrapper that immediately hands off processing to the 'ce-plan' skill.
Capability inventory: The agent is authorized to execute the local Python feedback script and perform planning tasks as delegated to 'ce-plan'.
Sanitization: The skill documentation explicitly mandates the redaction of secrets, tokens, and sensitive materials from all examples and artifacts produced during the planning process.

writing-plans