The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on various system and project-specific CLI tools to perform its functions.
Executes git commands (diff, status, log) and GitHub CLI (gh) to retrieve repository and Pull Request information.
Uses mmdc (Mermaid CLI) for diagram generation and python3 -m py_compile for code validation.
Invokes project-specific scripts such as flow-kanban, finalization.py, and merge_pipeline.py for workflow state transitions.
[DYNAMIC_CONTEXT_INJECTION]: Uses the dynamic context execution syntax to inject shell output into the agent's prompt at load time.
Specifically uses !git diff --name-only, !git status, and !git log --oneline -10 to provide the agent with current project state.
These commands are restricted to benign metadata gathering within the local repository.
[EXTERNAL_DOWNLOADS]: The research and prompt modules use WebSearch and WebFetch tools to retrieve external documentation and data.
This content is processed to generate research reports and inform code implementation decisions.
[PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection due to its processing of untrusted external data.
Ingestion points: Processes user-provided ticket files in .xml format and external content retrieved from the web.
Boundary markers: The provided file does not define explicit delimiters or instructions to ignore embedded commands in the ingested data.
Capability inventory: The skill has broad capabilities, including file writing (Write), shell command execution (Bash), and ticketing system modification (flow-kanban).
Sanitization: The skill references an external mechanism (readonly_session_guard.py) designed to restrict file writing capabilities during specific phases (research and review) to mitigate potential abuse.

workflow-wf