Skip to content
Back to skills

Sp Complete Design

ASecurity

Completion phase for a written plan. Self-review, gather missing input, update state, and request review — with explicit output-channel partitioning so only summaries and open questions reach the human.

  • 23 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added May 28, 2026
code-qualitygodebugging

Security analysis

A100/100

Scanned May 28, 2026

npx -y skills add tta-lab/ttal-cli --skill sp-complete-design --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Sp Complete Design?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Sp Complete Design
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/tta-lab-sp-complete-design/badge)](https://www.skillsdirectory.com/skills/tta-lab-sp-complete-design)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: sp-complete-design
description: "Completion phase for a written plan. Self-review, gather missing input, update state, and request review — with explicit output-channel partitioning so only summaries and open questions reach the human."
category: methodology
---

# Complete a Design Plan

## Overview

After a plan is written (by sp-planning or sp-debugging), this skill handles the final handoff phase: self-review, open questions, external research, state updates, and review routing.

Follow each step in order. Steps tagged **(→ persist)** write to state without surfacing to Neil. Steps tagged **(→ human)** output to Neil's context window.

---

## Step 1: Worker's-Eye Self-Review  (→ persist)

Re-read the plan as the worker who has to execute it. For every step ask:

- Could a worker execute this without asking?
- Are file paths, commands, and expected outputs exact?
- Are there hidden assumptions?

Note gaps to yourself only. Fix them in-place (edit the task tree / flicknote). No human-channel output from this step.

### Step 2: Identify Open Questions for Neil  (→ persist, then → human)

List decisions that genuinely need Neil's input (trade-offs you can't resolve, scope ambiguities, close architecture calls).

Write the list to persist first (annotate the task or append the orientation flicknote).

Then surface **only the questions** to the human channel — no reasoning, no alternatives, no self-review findings.

If there are no open questions, skip the human surface entirely.

### Step 3: Gather External Context If Needed  (→ persist)

If the plan needs cross-project reference, docs lookup, or web research, dispatch async:

```
ei ask --async "…" --project <alias>
ei ask --async "…" --web
```

Results land in `~/.einai/outputs/`. Fold findings back into the plan. Do not narrate the research to Neil.

### Step 4: Update the Plan with Review Findings and Answers  (→ persist)

Apply everything from steps 1–3 directly to the task tree / flicknote / annotations. Silent persist work:

```
task <uuid> plan
flicknote modify <id>
task <uuid> annotate '<note>'
```

### Step 5: Present the Summary  (→ human)

Surface only a concise summary of the finalized plan to Neil:

- One-paragraph what + why
- Bullet list of the major steps (not the details — those live in the tree)
- Any remaining risks or trade-offs

Keep it under ~200 words. No code blocks, no file paths, no step-by-step detail.

### Step 6: Request Review  (→ persist)

Run `ttal go <uuid>` to route the task to the plan reviewer. No human-channel output — the command itself is the handoff.

---

### Channel Discipline Checklist

- [ ] Self-review stayed in persist — no leakage to human channel
- [ ] Only open questions hit human in step 2 — no reasoning or alternatives surfaced
- [ ] Research dispatched async and folded into persist — no narrated findings to Neil
- [ ] Human-channel summary ≤200 words with no execution detail or file paths
- [ ] `ttal go <uuid>` ran cleanly in step 6

> **Reminder:** If persist-bound content leaked to the human channel, you burned Neil's context window.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.