The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Bash to execute standard development commands for progress tracking and synchronization. Evidence includes:
Usage of git status, git diff, and git log to verify implementation progress.
Usage of tail, jq, and grep to analyze local agent trace logs (.claude/state/agent-trace.jsonl) and the planning file.
These operations are limited to project metadata and are consistent with the skill's stated purpose of task management.
[EXTERNAL_DOWNLOADS]: The skill utilizes WebSearch during the planning phase to research technology stacks and architectures. This is an explicitly allowed tool in the frontmatter and is used to provide relevant suggestions to the user.
[DATA_EXFILTRATION]: While the skill reads internal agent trace data to determine which files have been edited, it does so locally to update the Plans.md file. There is no evidence of this data being transmitted to unauthorized external endpoints.
[PROMPT_INJECTION]: The skill contains functional constraints (e.g., "Do NOT load for: implementation, code review, or release tasks"), which are standard operational boundaries and do not constitute malicious prompt injection or safety bypass attempts.

harness-plan