Skip to content
Back to skills

Gl Grill Runner

ASecurity

Internal runner for the Decision GL gate.

  • 3 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added October 2, 2026
ai-agents

Works with

  • cli

Security analysis

A100/100

Scanned October 2, 2026

npx -y skills add lulufoo/lulu-workflow --skill gl-grill-runner --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Gl Grill Runner?

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

Security grade badge for Gl Grill Runner
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/lulufoo-gl-grill-runner/badge)](https://www.skillsdirectory.com/skills/lulufoo-gl-grill-runner)

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: decision/gl-grill-runner
description: Internal runner for the Decision GL gate.
meta-skill-version: 1.0.0
---

# gl-grill-runner

Surface decision-relevant user intent and critical uncertainties before direction
setting. Complete when the input is sufficient to enter E.

## Prerequisites

- Run `$GATE_CONTROL resolve-context`; pin `$CTX` (`active_gate` is `GL`).
- Apply `$CTX.domain_constraints` (`objective`, `role.instruction`,
  `domain.instruction`) to the dialogue.
- Probe questions: before the first probe in each GL entry, read
  `$SKILL_ROOT/shared/references/ask-protocol.md`; apply it to every probe.

## Must hold

Active X dimensions are preflight coverage labels, not diagnostic prompts.

| ID | Must hold |
|----|-----------|
| `G-direction-ready` | User intent is clear enough that further probing would not materially change the candidate direction set. |
| `G-diagnosis-preflight` | Decision-critical uncertainties that could overturn a direction are surfaced; each active X dimension has a conclusion or reasoned `na`. Unjustified all-`na` does not pass. |

## Ask domain

1. Anchor every probe to the locked Q problem + constraints.
2. Decision-domain intent only. No implementation interview, WBS, or unbounded
   plan grilling.
3. Freedom is which concrete question to ask — not any domain.

## Modes

| Mode | When | Behavior |
|------|------|----------|
| `probe` | Either Must hold row not met | Apply ask-protocol, then ask only the gap. May pick next lens; order not fixed. |
| `summarize` | Both Must hold rows met | Restate key intents once; ask if ready for E. |
| `close` | User confirms | `$GATE_CONTROL gate-close --gate GL --payload '<json>'` |

If the user rejects the summary: treat the denied point as a gap → `probe`.

## Signals

Apply Decision [Signals](../../SKILL.md#signals) throughout GL.

## Act

1. If `$CTX.gates.GL.status == stale`, follow `$DECISION_SKILL_DIR/references/rs-stale-gate-update.md`. Exactly one:
   - `$CTX.resume_gate` is `GL` → continue at 2.
   - `$CTX.resume_gate` is not `GL` → return `GATE_COMPLETE GL` and skip the loop.
2. Obtain locked Q via `$GATE_CONTROL get-payload` (or fields already on `$CTX`);
   do not start probes until Q payload is available.
3. Loop Modes (Signals as above) until `close` succeeds. On S1, register
   `source` is `GL`. Persist intents only via GL `gate-close` payload — do not
   dual-write exchanges to P.

## gate-close payload

```json
{
  "exchanges": [
    {
      "lens": "<active x_dimension_id>",
      "question": "<question>",
      "answer": "<user answer>",
      "na": false
    }
  ],
  "user_confirmed": true
}
```

- `lens` ∈ active `x_dimensions`; every active dim at least once (conclusion or `na: true`)
- Row rules and coverage: CLI (`dec_gate_control` / `--help`)
- `G-direction-ready` is framework-only (no extra payload slot)

## Exit

`GATE_COMPLETE GL` · `GATE_FAILED GL reason=<brief description>`

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.