The Agent Skills Directory

[SAFE]: The skill follows a robust operational model with defined lifecycle stages (Implementation, Spec Review, Quality Review, and Final Verification). It includes fail-fast logic to ensure inputs are valid and tasks are correctly identified before any modifications occur.
[COMMAND_EXECUTION]: The skill executes shell commands for implementing tasks and validating code changes. This capability is appropriate for its stated purpose as a development task execution tool and is governed by a requirement to provide documented evidence of all verification steps.
[PROMPT_INJECTION]: The skill ingests external data in the form of task descriptions and steps from a plan file, which creates a surface for indirect prompt injection. However, this risk is effectively mitigated by the skill's architecture, which requires separate reviewer subagents to validate work and the main agent to verify results using objective evidence.
Ingestion points: Ingests JSON input and reads task instructions from a provided plan_file absolute path.
Boundary markers: Not explicitly defined in the prompts to subagents.
Capability inventory: Filesystem read/write, git commit creation, and shell command execution.
Sanitization: No explicit sanitization or filtering of the content within the task steps is performed.

executing-single-task