The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes Python scripts located in the user's home directory (~/.claude/scripts/plan-manager.py and ~/.claude/scripts/adr-query.py) to manage task plans and Architecture Decision Records (ADRs). These executions are part of the intended administrative functionality for plan lifecycle management.
[COMMAND_EXECUTION]: The skill uses git commands (e.g., git status, git diff, git log, git rev-parse) to capture session state and track code changes for handoff purposes.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests and processes data from project files such as HANDOFF.json, .continue-here.md, and task_plan.md. These files are used to determine the next_action for the agent when resuming a session. However, the data is local to the project repository and used specifically for session continuity.
Ingestion points: HANDOFF.json, .continue-here.md, task_plan.md, and notes.md (read in references/resume.md and references/plan-files.md).
Boundary markers: None explicitly implemented to sanitize contents of these files before processing.
Capability inventory: The skill has access to Bash, Skill, and filesystem tools (Read, Write, Edit).
Sanitization: No explicit sanitization or validation of the next_action string from handoff files is mentioned before it is routed for execution.
[DATA_EXPOSURE]: The skill captures git logs and diffs to synthesize session summaries. While this exposes codebase history to the agent context, it is required for the stated purpose of session handoffs.

planning