The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as issue text and Markdown plans which could contain malicious instructions.
Ingestion points: Processes external inputs from source: <plan path | prompt | issue text> defined in the Execution Contract.
Boundary markers: Implements an explicit Gate 1: Contract Gate where the agent must summarize the task in a YAML-formatted execution_contract and seek human approval via AskUserQuestion before execution.
Capability inventory: The skill utilizes powerful tools including Bash, Write, Edit, and Task.
Sanitization: Employs a Fresh-Context Contract Review (Gate 3b) where a separate sub-agent (via the Task tool) verifies the implementation against the original contract in a clean environment.
[DYNAMIC_EXECUTION]: The skill uses the Task tool to spawn sub-agents for the purpose of code review and multi-pass execution (Test Pass vs Impl Pass).
Pattern: Uses a sub-agent to isolate the reviewer from the implementer to prevent self-evaluation bias.
Risk: While dynamic, this is a controlled architectural pattern used here to increase security and compliance.

dev-flow