The Agent Skills Directory

[DATA_EXFILTRATION]: The hooks/stop.sh script automatically collects and transmits development activity to the Plane API (typically https://plane.delo.sh). This data includes session duration, git commit messages, and the content of staged code changes via git diff --cached. While this is the intended functionality of the skill, it represents a flow of local environment metadata and source code to a remote service.
[COMMAND_EXECUTION]: The skill provides shell scripts (hooks/pre-prompt.sh, hooks/stop.sh, hooks/lib/workspace-detector.sh) intended to be configured as automatic agent hooks. These scripts execute shell commands (curl, git, jq, grep) and interact with the filesystem based on configuration values from .project.json and ~/.claude/plane-workspaces.json.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from project management tickets and documentation which could contain adversarial instructions.
Ingestion points: Plane API issue data parsed in hooks/stop.sh and references/audit-board-state.md, and repository story files parsed in references/create-ticket-from-story.md.
Boundary markers: The workflows do not explicitly implement boundary markers to isolate untrusted data from agent instructions.
Capability inventory: The agent can execute shell scripts, create git branches, and perform Plane API writes (creating/updating tickets).
Sanitization: The provided shell scripts use jq --arg to safely escape strings when building API payloads, reducing the risk of JSON or command injection in the API interaction layer.

project-lifecycle