The Agent Skills Directory

[PROMPT_INJECTION] (LOW): The skill ingests user-provided specifications to generate plans. This creates an indirect prompt injection surface where a malicious input could attempt to embed instructions into the generated tasks. This is a common characteristic of planning agents and is mitigated here by a strict output schema.
Ingestion points: User-provided specifications or requirements.
Boundary markers: Absent.
Capability inventory: Shell command generation (pytest, git), local file writing (docs/plans/), and handoff to external execution skills.
Sanitization: Absent.
[COMMAND_EXECUTION] (LOW): The skill generates plans containing standard development commands like pytest and git commit. These are contextually appropriate for a software engineering assistant and are intended for local execution within a version-controlled workspace.
[DATA_EXFILTRATION] (SAFE): No sensitive file access or network operations to external domains were detected. Operations are confined to the project's directory structure (e.g., docs/plans/, tests/).

writing-plans