Cheong43

Handle a small bounded Spwnr task.

Create an approved Spwnr plan before execution.

Shared Spwnr planning and orchestration rules.

Execute an approved Spwnr plan.

Audit Spwnr worker readiness.

Router for Spwnr commands.

Produce a concrete task result with rationale and residual risk instead of vague process commentary.

Collect repository facts, user constraints, and alternative approaches before recommending a direction.

Use when stable, reusable facts should be promoted into Layer 1 core knowledge from source code, docs, configuration, or verified project evidence.

Use on every run inside Mempedia so the agent behaves like an enterprise knowledge-base operator: search evidence first, update the correct memory layer, and treat skills as reusable operating procedures.

Use when the interaction is a transient event, greeting, one-off task update, or short-lived conversational episode that should stay in Layer 2 episodic memory.

Use for the independent post-turn memory agent that classifies each completed conversation into the four Mempedia layers before saving.

Use when the user expresses a stable preference, working style, formatting requirement, or persistent constraint that belongs in Layer 3 user preferences.

Use when the user asks for repository overview, README summary, architecture, project structure, module responsibilities, storage layout, or implementation inventory.

Use when a repeatable workflow should become a reusable skill document, and create or update a corresponding skills/<name>/SKILL.md file.

Run the mitosis-cli live planner replay harness and interpret the results. Use when testing branch-planner reliability, measuring fallback rates, comparing prompts under real model calls, checking for tool_call leakage, separating sandbox/network failures…