The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The skill uses standard shell commands (ls, find, tail, head) to search for checkpoint files within the local project directory structure (plans/reports/). These operations are safe and restricted to localized paths.
[PROMPT_INJECTION] (SAFE): The 'IMPORTANT Task Planning Notes' section contains standard instructions for task management and does not attempt to override system safety filters or extract restricted information.
[DATA_EXPOSURE] (SAFE): The skill reads internal workflow checkpoint files. It does not access sensitive system paths (like ~/.ssh), hardcoded credentials, or environment secrets.
[EXTERNAL_DOWNLOADS] (SAFE): No external dependencies, remote scripts, or network requests are initiated by this skill.
[Indirect Prompt Injection] (SAFE): This skill provides a surface for indirect prompt injection by design, as it ingests data from local checkpoint files to restore state.
Ingestion points: Local files matching the pattern plans/reports/memory-checkpoint-*.md.
Boundary markers: None; the skill relies on extracting structured JSON from the end of the file.
Capability inventory: Performs file listing and reading, and calls the TodoWrite tool to restore task state.
Sanitization: None; the skill is a meta-tool that assumes the integrity of its own previously saved state. Given the local scope and intended use-case for workflow management, this is considered a safe operational surface.

recover