The Agent Skills Directory

[COMMAND_EXECUTION]: The skill leverages a specific CLI utility (ssf) and standard Git tools to perform operations such as environment isolation, state management, and code reviewing. This involves executing shell commands with arguments derived from the project structure. Evidence includes the use of ssf isolate, ssf execution plan, ssf state transition, and git diff commands.
[INDIRECT_PROMPT_INJECTION]: The workflow involves reading and processing external documentation (e.g., specifications, design docs, and task lists) which is then interpolated into the instructions for subagents. This presents a risk where malicious instructions in these documents could influence the behavior of the implementer or reviewer agents. Ingestion points: Untrusted content is sourced from specs/, design.md, and tasks.md and placed into subagent templates via placeholders like [GLOBAL_CONSTRAINTS] and [BRIEF_FILE]. Boundary markers: The skill relies on Markdown structural markers rather than specialized delimiters or 'ignore' instructions for the ingested content. Capability inventory: The subagents have the capability to modify code, commit changes, and trigger further workflow steps through the ssf CLI. Sanitization: There is no evidence of content sanitization or verification within the skill's instructions.
[DYNAMIC_EXECUTION]: The skill uses templates to dynamically create subagent tasks at runtime. These templates define the behavior and constraints of implementer and reviewer agents, incorporating live data and context from the repository.

build-executor