The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses the platform's dynamic context syntax to execute a bundled Python script (history_manager.py status) at load time. This is a legitimate use of the feature to provide the AI agent with the current state of the workflow iterations from the local environment.
[INDIRECT_PROMPT_INJECTION]: The skill manages state for multi-step workflows, which inherently involves an indirect prompt injection surface as data produced in one phase (and stored in docs/history.json) influences the agent's context in later phases.
Ingestion points: The skill reads workflow state from docs/history.json and accepts status updates via command-line arguments.
Boundary markers: No specific boundary markers or instruction-override guards are present in the JSON storage logic.
Capability inventory: The script performs file reads, writes, and existence checks within the project's documentation directory.
Sanitization: Input phase names are validated against a hardcoded list of allowed phases (PHASES_ORDER), although specific identifiers (like phase_id) are interpolated into filenames without additional sanitization.

history-manager