agentleFS
nxxxsooonxxxsooo/loop/skills/deep-grill/SKILL.md

deep-grill

Skill2 starsChanged 7 days agoMITView on GitHub

Deeply grill an idea, decision, design, plan, or implied approach through evidence, failure modes, and user-owned decisions. Use discovery mode for raw ideas, clarification, "grill me", or an unfinished Product Brief; use audit mode for a formed target that needs a verdict or revision. Continue every reply in an active deep-grill session until its Product Brief or verdict is confirmed, without requiring the user to name the skill again.

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 1

    1. evolyob/AGY_SkillBook★ 1same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository12 other files it gives its agents.

The file

Show its rules and text

We split this skill into the rules it gives an agent

13 rules. About 0 are common, 11 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈11

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Run one evidence-driven branch loop in either discovery mode or audit mode.

    line 8
  2. Very specific:

    Use discovery mode when the input is a raw idea, the user asks to be interviewed or clarified, loop supplies work without a ready Product …

    line 12
  3. Distinctive:

    Use audit mode when the user supplies a formed plan, design, decision, or implied approach and wants autonomous pressure-testing, a …

    line 14
  4. Distinctive:

    If the mode is initially unclear, state and audit the smallest plausible approach.

    line 16
  5. Distinctive:

    After deep-grill starts, treat each answer to its question round as a continuation of the same session.

    line 20
  6. Distinctive:

    Ask the user only about frontier decisions that the agent cannot own.

    line 30
  7. Distinctive:

    Before presenting the first user decision, determine from the active client's tool metadata whether its native question interface is …

    line 34
  8. Distinctive:

    If callable, use one native question call for the round and include as many frontier decisions as the tool supports.

    line 36
  9. Distinctive:

    If unavailable in the current mode or unsupported, ask the same concise localized prose question immediately.

    line 37
  10. Distinctive:

    If the user explicitly chooses prose, keep the same frontier and recommendation structure in concise numbered prose.

    line 38
  11. Distinctive:

    When every material user-owned branch is settled, produce a concise Product Brief with the applicable fields:

    line 44
  12. Distinctive:

    Ask the user to confirm the brief through the native question interface when callable.

    line 58
  13. Distinctive:

    Do not implement or otherwise act on a Product Brief or verdict until the user confirms it or separately authorizes the next phase.

    line 64

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Deep Grill

Run one evidence-driven branch loop in either discovery mode or audit mode. Facts belong to the agent; goals, priorities, risk tolerance, taste, and authority belong to the user.

Select The Mode

Use discovery mode when the input is a raw idea, the user asks to be interviewed or clarified, loop supplies work without a ready Product Brief, or materially different approaches depend on a user-owned decision tree. Its output is a confirmed Product Brief.

Use audit mode when the user supplies a formed plan, design, decision, or implied approach and wants autonomous pressure-testing, a verdict, or revisions. Its output is an audit verdict.

If the mode is initially unclear, state and audit the smallest plausible approach. Switch to discovery mode only when continuing would invent a material user-owned decision. Discovery mode may use audit passes between question rounds; both modes are self-contained within deep-grill.

Keep The Session Active

After deep-grill starts, treat each answer to its question round as a continuation of the same session. Reconstruct the current branch tree from the conversation and artifacts, apply the answer, and continue. The user does not need to invoke deep-grill or loop again.

End the session when the user confirms the Product Brief or audit verdict, asks to stop, clearly changes tasks, or authorizes a separate next phase after deep-grill is complete.

Work The Branches

Map material branches in dependency order. The frontier is every branch whose prerequisites are settled.

Investigate factual branches with the available environment, sources, tests, and bounded experiments. For each provisional answer, use the strongest objection and a concrete failure scenario to keep, revise, or reject it. Mark insufficient evidence and provide a validation path instead of inventing facts. An unresolved factual check blocks only dependent branches.

Ask the user only about frontier decisions that the agent cannot own. Recompute the frontier after every answer or evidence change. Prioritize impact, uncertainty, and reversibility; stop at diminishing returns.

Use Native Questions

Before presenting the first user decision, determine from the active client's tool metadata whether its native question interface is callable now, unavailable in the current mode, or unsupported.

  • If callable, use one native question call for the round and include as many frontier decisions as the tool supports.
  • If unavailable in the current mode or unsupported, ask the same concise localized prose question immediately. Preserve the pending frontier and the recommended option, then continue from the ordinary reply.
  • If the user explicitly chooses prose, keep the same frontier and recommendation structure in concise numbered prose.

Each native question and prose fallback asks one decision, offers two or three mutually exclusive options, puts and marks the recommended option first, explains its main tradeoff in one sentence, and preserves free-form input. Do not repeat a successful native question call in Markdown.

Discovery Output

When every material user-owned branch is settled, produce a concise Product Brief with the applicable fields:

Product / offer:
Target user and context:
Problem and desired outcome:
Constraints and must-preserve behavior:
Non-goals:
Confirmed decisions:
Evidence and assumptions:
Success scenarios:
Residual risks:

Ask the user to confirm the brief through the native question interface when callable; otherwise use the same concise localized prose fallback under the same mode rules. The brief is ready for deep-design only after confirmation and when no root product decision remains open.

Audit Output

Report the target, verdict, recommended revisions, supporting evidence, strongest objections, failure scenarios, evidence limits, residual risks, unresolved items, and unexamined scope. Keep one or two bounded user choices inside the audit using the same native question and prose fallback rules. If a decision tree emerges, continue in discovery mode within deep-grill.

Do not implement or otherwise act on a Product Brief or verdict until the user confirms it or separately authorizes the next phase.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.