The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a limited set of file manipulation tools (Read, Write, Edit, Glob) for the express purpose of managing documentation within a .planify/ directory. It does not request or use shell access or arbitrary command execution.
[PROMPT_INJECTION]: The instructions include defensive constraints such as "Tu ne reposes aucune question, ne prends aucune décision, et n'inventes aucune tâche ni contrainte absente du récapitulatif," which strictly limits the agent's behavior to formatting existing data and prevents it from being easily redirected by embedded instructions in the source text.
[DATA_EXFILTRATION]: The skill has no network tools (allowed-tools configuration) and contains no external URL references, ensuring data remains within the local environment.
[SAFE]: Analysis of Indirect Prompt Injection (Category 8) surface reveals the following:
Ingestion points: Untrusted data enters the context via the conversation summary generated by the planify tool (referenced in SKILL.md).
Boundary markers: Explicit delimiters for the ingested data are not defined in the templates, though the system prompt provides negative constraints to prevent data invention.
Capability inventory: The skill uses Write and Edit tools to modify the local file system (SKILL.md).
Sanitization: No input sanitization or markdown escaping is mentioned in the formatting instructions.

planify-write-plan