The Agent Skills Directory

[SAFE]: The skill defines a structured software development lifecycle (SDLC) process. The tools requested (Read, Write, Bash, Task) are standard and necessary for an agent tasked with planning and executing code changes.
[INDIRECT_PROMPT_INJECTION]: The orchestrator ingests external user input to generate implementation plans and tasks which are subsequently executed by an agent with shell access.
Ingestion points: Initial context collection for the feature brief (SKILL.md workflow Step 1).
Boundary markers: None explicitly defined in the orchestration manifest to isolate user instructions from pipeline logic.
Capability inventory: The pipeline utilizes Bash, Write, Read, and Task tools during the execution and validation phases.
Sanitization: The skill effectively mitigates risk by implementing mandatory human-in-the-loop checkpoints at the direction review (ceo-review), architecture review (eng-review), and plan review (start) stages before any code is executed.

plan-pipeline