Index of every page on this site: llms.txt
All skills
paulrberg avatar

/grill-me

@c9e5a34
by Paul Bergpaulrberg/agent-skills106 stars
11
Run it yourself

Interview the user relentlessly about a plan or design until every branch of the decision tree has an explicit recorded decision and no open questions remain. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me".

Use this Skill: https://skilld.dev/gh/paulrberg/agent-skills/grill-me

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈66 tokens for metadata: the name and description. ≈754 when used: this file. ≈11 more on demand in 1 file.

Description uses 3.3% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈265 characters, or 3.3%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Grill Me

This skill is coordination-exempt: skip the ai-coord gate for its declared work.

Stress-test the plan in question rounds until every material branch is resolved and the user confirms shared understanding. User instructions take precedence over this skill.

Workflow

  1. Extract the current objective, constraints, assumptions, and already-made decisions. Map the design tree: each decision branches into dependent decisions. Investigate facts discoverable from the codebase, environment, or supplied evidence instead of asking the user. Put decisions to the user and wait for their answers.
  2. Maintain a running decision record with: decision, chosen option, rationale, dependencies, and consequences. Update it after every answer. Reopen a recorded decision only when new evidence conflicts with it or the user corrects it.
  3. When fact-finding benefits from delegation and subagents are available, delegate the investigation to a subagent that starts with a fresh context window. Otherwise, investigate directly. Treat pending fact-finding as an unsettled prerequisite. Defer only dependent questions while asking the rest of the currently answerable questions.
  4. Work in rounds. The frontier contains every unresolved decision whose prerequisites are settled. Ask the entire frontier in each round. A question that depends on another open question belongs in a later round. Number each question as ### 🔥 Decision <N> — <topic>, then give Recommended, Why, Tradeoff, and Question. Separate question cards with a horizontal rule. Word each Question so "yes" accepts its Recommended choice. Wait for the user's answers before the next round.
  5. After each answer, acknowledge ✅ Recorded: <choice> — <consequence>. Recompute the frontier after each round. When useful, show settled and currently open counts. Never show a percentage because the decision tree can grow.
  6. Continue through scope, users, interfaces, data/state, failure modes, security, operations, migration, testing, rollout, ownership, and stopping criteria only where they are material to this design.
  7. When no material open branches or required investigations remain, return ### ✅ Decision record with a decision/choice/rationale/consequence table. Then give ### ⚠️ Risks, accepted unknowns with owner/resolution point, rejected alternatives that matter, and ### 🏁 Completion criteria. Ask the user to explicitly confirm that the report reflects shared understanding. Wait for that confirmation before acting on the plan. If the user corrects the report, reopen the affected branches and continue the interview.

Completion requires every material branch to have a recorded decision or an explicitly accepted unknown with an owner and resolution point, followed by the user's explicit confirmation of shared understanding. The final report alone does not complete the workflow. After confirmation, act only within the user's authorized scope.

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No alerts7d3 checks · Risk SAFE
  • Gen Agent Trust Hub7d

    This skill facilitates a structured design review process by prompting the agent to ask critical questions about user plans. No security risks were identified in the implementation.

  • Socket10d

    No alerts

  • Snyk7d

    Risk: LOW · No issues

Signed by skilld at c9e5a34. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 3 days ago.

Activeupdated 3 weeks ago
coordination
exempt

README badge

README badge for paulrberg/agent-skills/grill-me