Structured deep-thinking framework for the moment before you act, not the act itself. Pick a mode (Solo or Interactive), emit the opening contract, run the 4-phase loop, and apply the mandatory stress-test at Tier Medium/High before committing.
When to use
Trigger on phrasings like "think deeply", "what should I do", "should we…", "help me decide", "walk me through this", "let's brainstorm".
Use when:
- the problem is underspecified, the right frame is unknown, or the user is stuck on framing
- multiple viable approaches have non-trivial tradeoffs (architecture, API shape, dependency choice)
- the change is expensive to reverse — schemas, public contracts, infra, security boundaries
- the codebase or domain is unfamiliar and the first-plausible-answer would be a guess
- root-cause reasoning is needed and there is no runtime to debug against (post-mortem, design review)
- a refactor could change behavior, boundaries, or invariants and you need to plan before touching code
- the user explicitly asks to brainstorm, walk through, explore options, or co-author a decision
- the task is likely to derail after the first obstacle and continuous execution discipline matters
Do NOT use when:
- runtime bug with a reproducible failure → use
do-debug(this skill is reasoning before execution;do-debugis iterative diagnosis with a runtime) - reviewing a PR or branch diff for merge → use
do-review - evaluating received review feedback → use
evaluate-code-review - the task is fully specified and only execution risk remains — just execute
- only summarizing, translating, or relaying already-known facts — no decision under ambiguity, no skill needed
Iron Laws (non-negotiable)
- Pick the mode FIRST. Default Solo for answer-or-execute requests. Switch to Interactive on a single trigger: user says "help me think"/"walk through"/"brainstorm"/"let's figure out", or stakeholder context is missing, or ≥2 viable options remain tied with no agent-owned tie-breaker, or Solo Phase C2 stress-test kills every candidate.
- Emit the opening contract on the first response (single line, machine-checkable):
WhereMode: <Solo|Interactive> Op: <op> Cynefin: <Clear|Complicated|Complex|Chaotic|Disorder> Tier: <Low|Medium|High>Op ∈ {SenseMaking, Extraction, Composition, Reshape, GroundedQA, WatchTrigger, Orchestration, SelfVerify}. Re-emit only if any field changes mid-session. - Ground before hypothesis. Observations and mechanisms separate from judgments at every phase.
- Match effort to Tier — see
references/foundations/effort-calibration.md. - Match thinking tools to Op —
references/foundations/operation-classification.mdgoverns Phase C1/C2 shape. - Keep ≥3 live options through Phase C only when Op = SenseMaking. Other ops have op-shaped C1 outputs (filled schemas, outlines, retrieved evidence, trigger specs, sequences, oracle readings).
- Stress-test before commit at Tier Medium or High. The trio is for SenseMaking; other ops use op-specific stress-test focus.
- End every session with a concrete next move + op-shaped verification check.
frameworks/is a library — load on demand from the routing table, never wholesale.- Never silently degrade Interactive into Solo. If the runtime can't fork, surface the limitation (
references/cross-runtime.md).
The 4-phase universal core loop
Phase A — Frame A1 classify domain · A2 classify operation · A3 reframe (if needed)
Phase B — Calibrate B1 effort tier · B2 minimum grounding (op-aware)
Phase C — Compare C1 op-specific output · C2 op-specific stress-test
Phase D — Commit D1 choose / produce · D2 verify (op-specific)
Phase A — Frame
- A1. Classify the domain (Cynefin) →
references/foundations/domain-classification.md. Write the domain in the opening contract. - A2. Classify the operation (8 cognitive operations) →
references/foundations/operation-classification.md. Write the operation in the opening contract. A2 selects the rest of Phase B and all of Phase C. - A3. Reframe the real question. Default: skip. Reframe only if A1 returned Disorder, or the user's framing fails the why-up test →
references/foundations/reframing.md(Abstraction Laddering).
Phase A exit: opening contract emitted with all four fields; reframe done if needed.
Phase B — Calibrate
- B1. Effort tier (Hard Choice × Confidence-vs-Quality) →
references/foundations/effort-calibration.md. Output: Low / Medium / High in the opening contract. - B2. Gather the minimum grounding set. Shape varies by Op; see
references/foundations/operation-classification.mdand the workflow file. Falsification ladder when evidence is the bottleneck →references/foundations/evidence-and-falsification.md.
Phase B exit: tier set; minimum op-shaped evidence collected.
Phase C — Compare
C1 and C2 are operation-specific. Run the workflow for the classified Op.
- C1. Produce the op-shaped output. Only SenseMaking requires ≥3 candidate verdicts; other ops produce schemas, outlines, invariant plans, evidence, trigger specs, sequences, or oracle readings.
- C2. Stress-test before committing at Tier Medium/High. SenseMaking → mandatory trio at
references/foundations/stress-test-trio.md; other ops use the op-specific C2 focus inoperation-classification.md.
Mandatory stress-test trio (SenseMaking only):
| # | Tool | Asks |
|---|---|---|
| 1 | Inversion (pre-mortem) | "Assume this fails in 6 months — what was the failure mode?" |
| 2 | Ladder of Inference rebuild | "What raw observation supports each judgment? Where did I leap?" |
| 3 | Second-order effects | "What does this break that's not on the page? Who else depends on this?" |
If any of the three kills every live option, escalate to Interactive (Iron Law #1). Do not silently pick the least-bad option.
Phase C exit: op-specific C1 output written AND op-specific C2 stress-test written. Outputs are written, not implied.
Phase D — Commit
- D1. Choose the next move (op-shaped): implement / produce / probe / ask one clarifying question / stop.
- D2. Verify or revise — every chosen path needs a verification check + a revision trigger. Shape varies by Op (verdict observation for SenseMaking; schema-validator pass for Extraction; audience-decision for Composition; oracle pass for SelfVerify; etc.). Full output contract:
references/foundations/output-contract.md.
Phase D exit: chosen path stated; op-shaped verification check written.
Mode mechanics
Solo mode (default)
Agent reasons alone, then acts. Anti-stall and escalation triggers in references/modes/solo-deep-think.md. Stay autonomous through blockers via references/workflows/continuous-execution.md.
Interactive mode (user in the loop)
6-step session with 5 forks. Full mechanics: references/modes/interactive-brainstorm.md. Fork discipline, one-question-at-a-time, YAGNI, pushback: references/interaction-patterns.md.
| Step | Phase | What runs | Fork |
|---|---|---|---|
| 1 — Classify | A1 | Cynefin classifier (3 questions) | Fork 1: domain correct? |
| 2 — Decompose | A2 (extended) | references/frameworks/decomposition-tools.md or references/foundations/reframing.md |
Fork 2: decomposition captures the problem? |
| 3 — Explore | C1 | references/frameworks/six-thinking-hats.md / references/frameworks/first-principles.md / references/frameworks/zwicky-box.md / references/frameworks/systems-tools.md |
Fork 3: options resonate? |
| 4 — Evaluate | B2 + scoring | Hard Choice classifier → references/frameworks/decision-matrix.md (Tier set in Step 1) |
Fork 4: factors + weights right? |
| 5 — Stress-test | C2 | Op-specific stress-test from references/foundations/operation-classification.md (use the SenseMaking trio in references/foundations/stress-test-trio.md when Op: SenseMaking) |
Fork 5: blind spots change the pick? |
| 6 — Communicate | D | references/foundations/output-contract.md (10-section deliverable) |
(none) |
If the Interactive artifact would exceed 3,000 words, re-scope to the highest-risk sub-problem before producing.
Solo routing — minimum read set by Op
Files listed are the minimum; add foundations as Tier escalates.
| Situation (by Op) | Read this set |
|---|---|
| Generic Sense-Making (research, judgment, evaluation) | references/modes/solo-deep-think.md, references/foundations/operation-classification.md, references/workflows/sense-making.md, references/foundations/stress-test-trio.md |
Ambiguous / post-do-debug runtime hypothesis (SenseMaking, bug variant) |
references/workflows/bug-tracing.md, references/foundations/evidence-and-falsification.md |
| Recurring / "we already fixed this" (SenseMaking, systemic) | references/workflows/recurring-issue.md |
| Structured Extraction (mess → schema) | references/workflows/structured-extraction.md |
| Generative Composition (artifact, doc, code feature) | references/workflows/generative-composition.md |
| Reshape & Repurpose — code (refactor with invariants) | references/workflows/refactor-thinking.md |
| Reshape & Repurpose — text/data | references/foundations/operation-classification.md (Op 4 inline) |
| Grounded Q&A (corpus answer with citations) | references/workflows/grounded-qa.md |
| Watch & Trigger (design the trigger conditions) | references/foundations/operation-classification.md (Op 6 inline) |
| Cross-System Orchestration (sequence design) | references/foundations/operation-classification.md (Op 7 inline) |
| Iterative Self-Verification (write → test → fix with oracle) | references/workflows/iterative-self-verification.md |
| Planning a complex task (any Op) | references/workflows/task-planning.md, references/foundations/stepwise-reasoning.md |
| Staying autonomous through blockers (any Op) | references/workflows/continuous-execution.md |
| Especially high-stakes / hard-to-reverse (any Op) | + references/foundations/ultrathinking.md |
Output contracts
- Solo SenseMaking — Minto Pyramid: first sentence = verdict, body = 3–5 evidence-backed arguments, last sentence = verification check.
- Solo non-SenseMaking ops — use the workflow file's output contract: Extraction schema/provenance, Composition artifact/provenance, Reshape diff/invariants, GroundedQA answer/citations, WatchTrigger trigger spec, Orchestration sequence, or SelfVerify oracle log.
- Interactive mode — 10-section contract: Approach / Problem shape / Decomposition / Options explored / Evaluation / Assumptions / Blind spots / Second-order effects / Ranked summary / Recommended next step.
Full spec: references/foundations/output-contract.md.
Voice discipline (literal forms)
Mode: <Solo|Interactive>Op: <SenseMaking|Extraction|Composition|Reshape|GroundedQA|WatchTrigger|Orchestration|SelfVerify>Cynefin: <domain>·Tier: <Low|Medium|High>Live options: 1) … 2) … 3) …(in C1, only whenOp: SenseMaking)Filled schema:/Outline:/Retrieved evidence:/Trigger spec:/Sequence:/Oracle reading:(in C1, op-specific)Pre-mortem failure mode: …(in C2, only whenOp: SenseMaking)Coverage: …/Form-substance match: …/Hallucination scan: …/ etc. (in C2, op-specific)Verification check: …(in D2)
Session completion contract
Every finished session states:
- chosen next move or produced artifact
- assumptions that matter
- risks/blind spots, plus rollback when
Tier: High - op-shaped verification check
- handoff target when another skill should continue
Anti-patterns
- Acting on the first plausible explanation
- Skipping Phase A1 (Cynefin) because the prompt feels straightforward
- Skipping Phase A2 (Op classification) and defaulting to SenseMaking for everything
- Running the SenseMaking stress-test trio on a non-SenseMaking operation
- Forcing ≥3 options on a task that isn't SenseMaking (Extraction wants completeness, Composition wants outline-fit)
- Letting one attractive option kill comparison too early (in SenseMaking)
- Treating Phase C2 as ceremonial when it's load-bearing
- Stopping after the first obstacle when better local moves still exist
- Claiming success before stating the (op-shaped) verification result
- Silently degrading Interactive into Solo
- Reading all
frameworks/*, or loading more than one decomposition + one generation + one evaluation/stress tool in a single pass - Asking the user a clarifying question mid-Solo (you've hit an escalation trigger — switch to Interactive instead)
Rationalization checks
- If classification, stress-test, or verification feels obvious, write it anyway.
- High stakes raise Tier; Interactive still needs a mode gate.
- The trio is SenseMaking-only; other ops use op-shaped C2 checks.
- If every candidate dies, switch modes or hand off. Do not pick the least-bad path silently.
Escalation gates
- Solo → Interactive when: all candidates/attempts die in C2 stress-test, or user-flagged, or auto-detected (missing stakeholder context, ≥2 tied options).
- Interactive → Solo never silently — only on explicit user request.
- Out to
do-debugwhen: the task is a runtime bug with a reproducible failure, or ado-thinksession produces a testable runtime hypothesis. - Out to
check-completionwhen: declared "done" but unsure what's actually verified. - Do NOT recommend
do-thinkas the next-step (no infinite regress). If more thinking is needed, name the sub-topic and explicitly hand off.
do-debug route-back handoff
When this session — especially one entered from do-debug after 3 failed fixes — produces a falsifiable runtime hypothesis, hand back to do-debug with: symptom card / observed failure, failed-fix summary if known, selected hypothesis + mechanism, falsification experiment, narrow next diagnostic or fix boundary, and verification check.
Reference routing — master table
Foundations are self-sufficient. frameworks/ is a library accessed from this table, Interactive mode, or an operation workflow that explicitly names the framework.
Foundations (load-bearing primitives)
| File | Read when |
|---|---|
references/foundations/core-loop.md |
Always — the 4-phase loop with opening contract |
references/foundations/domain-classification.md |
Phase A1 — Cynefin classifier |
references/foundations/operation-classification.md |
Phase A2 — the 8 cognitive operations + per-op routing |
references/foundations/reframing.md |
Phase A3 — Abstraction Laddering when framing is wrong or A1 returned Disorder |
references/foundations/effort-calibration.md |
Phase B1 — Hard Choice × Confidence-vs-Quality |
references/foundations/evidence-and-falsification.md |
Phase B2 + Phase C2 — evidence ladder, falsification, Ladder of Inference rebuild |
references/foundations/stepwise-reasoning.md |
When tempted to leap from symptom to cause |
references/foundations/stress-test-trio.md |
Phase C2 for SenseMaking — Inversion + Ladder + Second-Order |
references/foundations/ultrathinking.md |
Tier High — irreversible / one-shot / broad blast radius |
references/foundations/output-contract.md |
Always — SenseMaking Minto, op-shaped Solo contracts, 10-section Interactive |
Modes
| File | Read when |
|---|---|
references/modes/solo-deep-think.md |
Mode = Solo — autonomous loop, anti-stall, escalation triggers |
references/modes/interactive-brainstorm.md |
Mode = Interactive — 6-step + 5-fork mechanics |
Workflows (operation-specific recipes)
| File | Op | Read when |
|---|---|---|
references/workflows/sense-making.md |
SenseMaking | Generic research / judgment / evaluation |
references/workflows/bug-tracing.md |
SenseMaking (bug variant) | Ambiguous / post-do-debug runtime hypothesis framing |
references/workflows/recurring-issue.md |
SenseMaking (systemic) | "We already fixed this and it's back" — Iceberg + Connection Circles |
references/workflows/structured-extraction.md |
Extraction | Mess → schema (transcript → todos, invoice → ledger) |
references/workflows/generative-composition.md |
Composition | Context + form → artifact (cover letter, RFP, deck, code feature) |
references/workflows/refactor-thinking.md |
Reshape (code) | A refactor where invariants must survive |
references/workflows/grounded-qa.md |
GroundedQA | Question + corpus → cited answer |
references/workflows/iterative-self-verification.md |
SelfVerify | Write → test → fix loop with deterministic oracle |
references/workflows/task-planning.md |
(universal) | A large or fuzzy task that needs slicing |
references/workflows/continuous-execution.md |
(universal) | Need to keep moving without frequent human checkpoint |
Frameworks (selection-capped library)
Selection rule: at most one decomposition tool, one generation tool, and one evaluation/stress tool per pass. In Solo, load no framework unless C1 generation or C2 evaluation needs it. In Interactive, Steps 2–5 each pick one primary framework.
| File | Read when |
|---|---|
references/frameworks/six-thinking-hats.md |
Phase C1 needs perspective rotation; Interactive Step 3 |
references/frameworks/zwicky-box.md |
Phase C1 generator is dry — combinatorial generation |
references/frameworks/first-principles.md |
Analogies are failing; suspect inherited assumptions |
references/frameworks/decomposition-tools.md |
Issue Trees / Ishikawa for problem decomposition |
references/frameworks/decision-matrix.md |
Phase C2 with ≥3 options + multiple weighted factors; Hard Choice classifier |
references/frameworks/systems-tools.md |
Connection Circles + Reinforcing/Balancing Loops — feedback dynamics |
references/frameworks/productive-thinking-drive.md |
Defining success criteria with the DRIVE check |
references/frameworks/interpersonal-tools.md |
SBI for feedback content; Conflict Resolution Diagram for stakeholder conflict |
Cross-cutting (Interactive infrastructure)
| File | Read when |
|---|---|
references/interaction-patterns.md |
Interactive mode — fork discipline, one-question-at-a-time, YAGNI, pushback |
references/cross-runtime.md |
Interactive on a non-Claude runtime — ask-user tool lookup + prose fallback |
Final test
Before acting, ask: Do I know enough to make the next move safer and clearer than acting immediately would be? If not, gather the smallest missing evidence first.