Run completely UNDER CONTROL — execute the SINGLE next Beads task through the FULL task engine (understand → map → plan-check → parallel subagents → TDD → checks → reviewers → verifier → evaluator → debug-on-fail → commit → close), in this session, showing every step and pausing at human gates. One task done excellently, then stop. control = one observed step of auto, same engine, no cuts.
5 stars
0 votes
0 copies
1 view
Added June 9, 2026
ai-agentsgocode-reviewgitsecurity
Security analysis
A100/100
Scanned June 9, 2026
$npx -y skills add 23ag1/completely --skill control --agent claude-code
Installs into .claude/skills of the current project.
Are you the author of Control?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/23ag1-control)
---
name: completely:control
description: Run completely UNDER CONTROL — execute the SINGLE next Beads task through the FULL task engine (understand → map → plan-check → parallel subagents → TDD → checks → reviewers → verifier → evaluator → debug-on-fail → commit → close), in this session, showing every step and pausing at human gates. One task done excellently, then stop. control = one observed step of auto, same engine, no cuts.
version: 0.7.0
user-invocable: true
---
Execute the **single next `bd ready` task** — exactly ONE — through the full task engine
(`core/task-engine.md`), in this session, with maximum observation and control.
Follow steps 0–10 of the engine for that one task: claim → understand (spawn gsd-codebase-mapper /
gsd-phase-researcher as needed) → plan-check (goal-backward) → decompose & spawn subagents in
parallel where write-zones are disjoint (serialize with `bd merge-slot`) → TDD + craft skills →
`cmpl check` + `cmpl lint` → spawn code-reviewer + security-reviewer → gsd-verifier + the
default-FAIL evaluator → gsd-debugger on repeated failure → **LAND: bd comment evidence → git commit
(confirm it landed) → ONLY THEN `bd close` if the evaluator ACCEPTED**.
Show me each subagent's result. PAUSE at any human gate (checkpoint) and at STOP-conditions. After
the one task lands, STOP and report — I decide whether to run the next.
Identical to one iteration of `/completely:auto` — same full engine — just observed and
one-at-a-time. Nothing is reduced.
