tk-grill

Decision Grilling

Start only through an explicit /tk-grill, $tk-grill, or host skill selection. Do not invoke automatically when the user asks for a quick opinion, ordinary brainstorming, repository explanation, or implementation planning.

Boundary

Own the interview from a rough idea, decision, or plan through confirmed shared understanding. Keep all state in the current conversation.

Do not create or modify source, tests, config, Git state, tickets, Seed, SDD, domain artifacts, general scratch files, ledgers, or task graphs. The Visual questions branch alone permits one HTML comparison under .tigerkit/tmp/tk-grill/<run-id>/ and its required ignore setup. Do not invoke another skill or begin implementation automatically. After synthesis, offer only relevant optional routes:

• tk-prep when the result has become an implementation request or ticket;
• tk-domain when the user wants repository-native vocabulary or a durable decision;
• stop when no further action is needed.

For a user-only next skill, give one line with /<skill> and say that the current conversation carries over. Supply a full handoff prompt only for a different session; respect disable-model-invocation without attempting the call or simulating its procedure.

Decision tree and frontier

Map the subject as a decision tree. A branch may become answerable only after its prerequisites are settled. The current frontier is every unresolved decision that can be answered now without guessing at another unresolved answer.

Work in rounds:

1. Separate established facts, unresolved assumptions, and user-owned decisions.
2. Resolve safely researchable facts from the conversation, supplied files, repository, connected tools, or bounded public sources. Do not ask the user for a fact that available evidence can answer. Bind shell lookups to the verified repository root; quote search globs (for example rg --glob '*.ts' or grep --include='*.ts') and omit separator-only commands, especially unquoted words starting with = in zsh.
3. Ask the whole current frontier in one round. Group only independent questions; never impose a fixed question count or round size. Before the first question round in this turn, read question rounds and apply its format to every round. If any frontier question compares layout, hierarchy, density or navigation alternatives, attach the visual question aid in this same round.
4. Defer a question when its answer depends on an unresolved prerequisite. An unresolved research fact blocks only its dependent branch, not the rest of the frontier.
5. After the user answers, update the tree, identify contradictions or newly exposed assumptions, recompute the frontier, and continue.

Do not turn fact-finding into unbounded research. Investigate only facts that can materially change the current decision tree.

For a consequential choice between alternatives, test the recommendation against its decisive assumptions and priorities: identify an evidence-supported change in conditions that would favor another option, when one is known. Research checkable constraints rather than asking the user; ask only for user-owned priorities. If the turning condition is unknown, retain that uncertainty rather than inventing a threshold. Skip this analysis for minor, easily reversible choices.

Visual questions

For unresolved spatial alternatives, read visual questions and attach one HTML comparison to the numbered text round. Keep every decision complete in chat; the aid also collects nonvisual choices from that same round. The Boundary permits only this question aid and its canonical artifact setup.

Question contract

Goals, priorities, trade-off preferences, scope, meaning, tone, and accepted risk belong to the user. Never silently choose one. When a useful recommendation is possible, include the recommendation and its reason instead of presenting neutral options and delegating all judgment.

When a recommendation depends on an existing mechanism or on when data is present, cite the path:line that establishes it; otherwise investigate before recommending.

Do not ask downstream questions merely to appear thorough. Relentless means leaving no silent decision branch, not maximizing question volume.

Completion

When the frontier appears empty:

1. Check the whole tree for an unvisited branch, hidden assumption, contradiction, or intentionally open risk.
2. Summarize the current understanding briefly and ask the user to confirm that it is shared and complete.
3. Do not act on the result before confirmation. If the user corrects it, update the tree and continue from the new frontier.
4. After confirmation, return a compact synthesis of the purpose or conclusion, key decisions and reasons, intentionally open assumptions or risks, and optional next routes. Include a material turning condition when identified. Create no additional artifact; keep only an already-authorized visual question aid.

Source

Adapted the decision-tree, prerequisite-aware frontier, whole-frontier rounds, fact-versus-decision ownership, recommendations, and shared-understanding gate from mattpocock/skills skills/productivity/grilling at commit 6654f6b60cd9d5be8b54c6fafe44346dabeb3b76.

TigerKit removes mandatory subagent runtime assumptions, wrappers such as grill-me and grill-with-docs, automatic implementation transition, and persistent state. The original copyright and MIT notice are in LICENSE.txt.

Artifact Paths

Artifacts require task authority. Before writes/checkout/transport/ignore setup, read artifact paths for Git exclusion/path/ownership checks. Reuse identical in-context content; hash-check uncertainty and reread changes/loss; always recheck safety. Use .tigerkit/ unless a destination is explicit or the eligible durable-root branch applies. Chat-only: skip file/ignore setup and this reference. This grants no unrelated mutation/publication authority. Prefer fresh directories and supported trash for authorized non-secret cleanup; target only recorded literal resolved absolute run-owned paths, never cd plus glob/relative deletion.

Reader-facing Result Summary

Return a brief result with the outcome, actual verification, material limitations and direct deliverable links. Distinguish plans/skipped checks from executed work; retain outstanding manual checks and real cleanup, blocker or human-action details. Never claim unverified completion or cleanup. Preserve the owner's required artifact, publication preview, approval boundary, machine return, exact status/footer and final question block; keep nested returns compact and omit empty sections.

Output Notation

Use ASCII numbering (1. Item, (1) Item) and - Item, not enclosed/keycap numerals. Show complete absolute clickable file paths, including examples, without ellipses. Preserve exact code, commands, identifiers, URLs, quotations and verified UI labels unless their edit is authorized.

For chat comparisons with multiple shared attributes, use narrow GFM pipe tables (about four columns, header, short cells); otherwise use lists. No box-drawing, fenced or empty tables. Owner-specific previews take precedence; publication bodies and file artifacts retain their own table formats.

In rendered Markdown prose, use hyphens for ranges and escape literal tildes as \~; preserve exact code, quotations and identifiers.

For authorized temporary input files, follow Artifact Paths before creation and consumption; a prefilled template is not completed input. This grants no write authority.

User Questions

Investigate facts; reuse authorization. For user-owned decisions read question rounds; otherwise proceed. Plain chat, even one approval: standalone ---, then ❓ **Q<n> · <title>**: <question> and ➡️ <recommendation> per item. Ask only settled-prerequisite questions; separate approval. Questions come last; put status, risks and plans before them.

Skill Feedback

For a reusable incident, retain minimal non-secret evidence in the current conversation and suggest tk-retro; do not write it to persistent memory unless the user asks. Invocation, feedback artifacts, installed-skill edits and publication require separate authorization through the owning workflow; this pointer grants none.