{"id":"e55f068b-4f06-4549-b59b-4e1698e6ea1d","shortId":"PKFm4v","kind":"skill","title":"deepthink","tagline":"Use when a problem resists quick answers, debugging stalls, analysis feels shallow, confidence is low, hypotheses are competing, reasoning loops repeat, or a hard problem needs evidence tracking.","description":"# Deepthink\n\nStructured extended reasoning with hypothesis tracking and confidence progression. Prevents circular thinking by explicitly tracking what's been explored, what evidence exists, and what confidence level has been reached.\n\nReferences `perspectives` for multi-angle evaluation when confidence is stuck and a fresh frame is needed.\n\n<workflow>\n\n## Workflow\n\n### 1. Frame the Problem\n\nState what you're trying to understand or decide, specifically. Vague framing produces vague investigation.\n\n### 2. Form Initial Hypothesis\n\nYour best guess based on available information. One sentence. Confidence: `exploring`.\n\nDon't skip this step — even a weak hypothesis focuses investigation better than no hypothesis.\n\n### 3. Gather Evidence\n\nRead code, check docs, run tests, trace execution. Record what you find at each step. Every piece of evidence should be evaluated against the current hypothesis.\n\n> **Note:** Sequential-thinking or similar extended reasoning tools can complement complex sub-steps within this workflow — particularly during evidence gathering (step 3) or when evaluating a hypothesis with many interdependencies. Use them to decompose a stuck sub-step without abandoning the overall hypothesis tracking structure.\n\n### 4. Evaluate Against Hypothesis\n\nDoes the evidence support, contradict, or require revision?\n\n- **Supports:** confirms a specific aspect of the hypothesis\n- **Contradicts:** rules out a specific aspect, requiring revision\n- **Requires revision:** the hypothesis was wrong in some way — update it now\n\n### 5. Update Confidence\n\nBased on evidence quality and coverage, update the confidence level:\n\n| Level | Meaning | Action |\n| ----- | ------- | ------ |\n| `exploring` | Just started, no hypothesis yet | Gather initial evidence, form hypothesis |\n| `low` | Have a hypothesis but weak evidence | Seek confirming/disconfirming evidence |\n| `medium` | Evidence supports hypothesis but gaps remain | Fill specific gaps, check edge cases |\n| `high` | Strong evidence, minor uncertainties | Verify the uncertainties aren't critical |\n| `certain` | Conclusive evidence, ready to act | Synthesize findings and present |\n\n**Escalation rule:** If confidence has not increased after 3 investigation steps, stop and reassess. Either the hypothesis is too broad, you're looking in the wrong place, or you need a different tool (`flow:tracer`, `flow:perspectives`).\n\n### 6. Decide: Continue or Conclude\n\n- **Continue:** identify exactly what's missing and loop back to step 3 with a specific target\n- **Conclude:** if confidence is `high` or `certain`, synthesize findings and present\n\nInvestigation is complete when: confidence is `high`/`certain`, all evidence-against items are explained, the hypothesis is a specific actionable conclusion, and unexplored areas are evaluated or ruled out as non-critical.\n\n</workflow>\n\n<guardrails>\n\n### Guardrails\n\n- **Evidence hoarding** — Reading files without updating hypothesis. Every read should confirm, contradict, or refine your current hypothesis. If it doesn't, you're reading the wrong thing.\n- **Premature conclusion** — Presenting hypothesis as conclusion before gathering evidence. A hypothesis is not a conclusion. Don't present it as one.\n- **Permanent exploring** — Staying at `exploring` after 5+ checks without narrowing. Formulate a hypothesis and commit to testing it.\n- **Circular investigation** — Revisiting same evidence without new framing. If you're back where you started, the hypothesis needs to change, not the evidence gathering.\n\n</guardrails>\n\n<validation>\n\n### Validation Checkpoint\n\nBefore presenting the conclusion, verify:\n\n- [ ] Hypothesis was updated at least once during investigation\n- [ ] Evidence for AND against was recorded\n- [ ] Confidence level progression is justified by evidence\n- [ ] Investigation concluded with a specific, actionable finding\n\n</validation>\n\n<example>\n\n## Example\n\n**Debugging:** \"Tests pass locally but fail in CI.\"\n\n- **Hypothesis 1:** \"Flaky test — timing issue\" (confidence: `exploring`)\n- **Evidence:** CI logs show deterministic failure on same test. Rules out flakiness. → Revise.\n- **Hypothesis 2:** \"Environment difference — missing env var\" (confidence: `low`)\n- **Evidence:** Compared CI env vs local. Found `DATABASE_URL` uses different host. Test creates real DB connection. → confidence: `medium`.\n- **Hypothesis 3:** \"CI database not seeded\" (confidence: `medium`)\n- **Evidence:** CI setup script skips seed step for test DB. Local has leftover seed data. → confidence: `high`.\n- **Conclusion:** CI test DB is empty. Fix: add seed step to CI pipeline before tests.\n\n</example>\n\n## Complements\n\n- **systematic-debugging** — deepthink provides structured hypothesis evolution when debugging stalls after 3+ iterations\n- **brainstorming** — deepthink enables deeper analysis during the design phase when approaches need thorough evaluation\n- **flow-plan** — deepthink supports thorough requirement analysis for complex decomposition decisions\n\n## References Index\n\n- **[Reasoning Strategy](references/reasoning-strategy.md)** — When to use deepthink, the 6-step workflow, and anti-patterns to avoid\n- **[Confidence Tracking](references/confidence-tracking.md)** — Confidence levels table, what to track at each step, escalation rule, and completion criteria","tags":["deepthink","flow","cofin","agent-skills","ai-agents","beads","claude-code","codex","cursor","developer-tools","gemini-cli","opencode"],"capabilities":["skill","source-cofin","skill-deepthink","topic-agent-skills","topic-ai-agents","topic-beads","topic-claude-code","topic-codex","topic-cursor","topic-developer-tools","topic-gemini-cli","topic-opencode","topic-plugin","topic-slash-commands","topic-spec-driven-development"],"categories":["flow"],"synonyms":[],"warnings":[],"endpointUrl":"https://skills.sh/cofin/flow/deepthink","protocol":"skill","transport":"skills-sh","auth":{"type":"none","details":{"cli":"npx skills add cofin/flow","source_repo":"https://github.com/cofin/flow","install_from":"skills.sh"}},"qualityScore":"0.455","qualityRationale":"deterministic score 0.46 from registry signals: · indexed on github topic:agent-skills · 11 github stars · SKILL.md body (5,308 chars)","verified":false,"liveness":"unknown","lastLivenessCheck":null,"agentReviews":{"count":0,"score_avg":null,"cost_usd_avg":null,"success_rate":null,"latency_p50_ms":null,"narrative_summary":null,"summary_updated_at":null},"enrichmentModel":"deterministic:skill-github:v1","enrichmentVersion":1,"enrichedAt":"2026-05-18T19:07:36.078Z","embedding":null,"createdAt":"2026-04-23T13:03:58.645Z","updatedAt":"2026-05-18T19:07:36.078Z","lastSeenAt":"2026-05-18T19:07:36.078Z","tsv":"'1':78,554 '2':97,575 '3':127,179,323,368,603,655 '4':204 '5':244,473 '6':352,693 'abandon':198 'act':310 'action':259,404,542 'add':634 'analysi':11,661,678 'angl':65 'answer':8 'anti':698 'anti-pattern':697 'approach':667 'area':408 'aren':302 'aspect':220,229 'avail':106 'avoid':701 'back':365,496 'base':104,247 'best':102 'better':123 'brainstorm':657 'broad':334 'case':293 'certain':305,379,391 'chang':504 'check':132,291,474 'checkpoint':510 'ci':552,562,585,604,611,628,638 'circular':41,485 'code':131 'commit':481 'compar':584 'compet':19 'complement':166,642 'complet':386,717 'complex':167,680 'conclud':356,373,538 'conclus':306,405,447,451,460,514,627 'confid':14,38,55,68,110,246,255,318,375,388,530,559,581,600,608,625,702,705 'confirm':217,429 'confirming/disconfirming':279 'connect':599 'continu':354,357 'contradict':212,224,430 'coverag':252 'creat':596 'criteria':718 'critic':304,417 'current':154,434 'data':624 'databas':590,605 'db':598,619,630 'debug':9,545,645,652 'decid':90,353 'decis':682 'decompos':191 'decomposit':681 'deeper':660 'deepthink':1,30,646,658,674,691 'design':664 'determinist':565 'differ':346,577,593 'doc':133 'doesn':438 'edg':292 'either':329 'empti':632 'enabl':659 'env':579,586 'environ':576 'escal':315,714 'evalu':66,151,182,205,410,670 'even':117 'everi':145,426 'evid':28,51,129,148,176,210,249,268,277,280,282,296,307,394,419,454,489,507,524,536,561,583,610 'evidence-against':393 'evolut':650 'exact':359 'exampl':544 'execut':137 'exist':52 'explain':398 'explicit':44 'explor':49,111,260,468,471,560 'extend':32,162 'fail':550 'failur':566 'feel':12 'file':422 'fill':288 'find':141,312,381,543 'fix':633 'flaki':555,572 'flow':348,350,672 'flow-plan':671 'focus':121 'form':98,269 'formul':477 'found':589 'frame':74,79,93,492 'fresh':73 'gap':286,290 'gather':128,177,266,453,508 'guardrail':418 'guess':103 'hard':25 'high':294,377,390,626 'hoard':420 'host':594 'hypothes':17 'hypothesi':35,100,120,126,155,184,201,207,223,235,264,270,274,284,331,400,425,435,449,456,479,501,516,553,574,602,649 'identifi':358 'increas':321 'index':684 'inform':107 'initi':99,267 'interdepend':187 'investig':96,122,324,384,486,523,537 'issu':558 'item':396 'iter':656 'justifi':534 'least':520 'leftov':622 'level':56,256,257,531,706 'local':548,588,620 'log':563 'look':337 'loop':21,364 'low':16,271,582 'mani':186 'mean':258 'medium':281,601,609 'minor':297 'miss':362,578 'multi':64 'multi-angl':63 'narrow':476 'need':27,76,344,502,668 'new':491 'non':416 'non-crit':415 'note':156 'one':108,466 'overal':200 'particular':174 'pass':547 'pattern':699 'perman':467 'perspect':61,351 'phase':665 'piec':146 'pipelin':639 'place':341 'plan':673 'prematur':446 'present':314,383,448,463,512 'prevent':40 'problem':5,26,81 'produc':94 'progress':39,532 'provid':647 'qualiti':250 'quick':7 're':85,336,441,495 'reach':59 'read':130,421,427,442 'readi':308 'real':597 'reason':20,33,163,685 'reassess':328 'record':138,529 'refer':60,683 'references/confidence-tracking.md':704 'references/reasoning-strategy.md':687 'refin':432 'remain':287 'repeat':22 'requir':214,230,232,677 'resist':6 'revis':215,231,233,573 'revisit':487 'rule':225,316,412,570,715 'run':134 'script':613 'seed':607,615,623,635 'seek':278 'sentenc':109 'sequenti':158 'sequential-think':157 'setup':612 'shallow':13 'show':564 'similar':161 'skill' 'skill-deepthink' 'skip':114,614 'source-cofin' 'specif':91,219,228,289,371,403,541 'stall':10,653 'start':262,499 'state':82 'stay':469 'step':116,144,170,178,196,325,367,616,636,694,713 'stop':326 'strategi':686 'strong':295 'structur':31,203,648 'stuck':70,193 'sub':169,195 'sub-step':168,194 'support':211,216,283,675 'synthes':311,380 'systemat':644 'systematic-debug':643 'tabl':707 'target':372 'test':135,483,546,556,569,595,618,629,641 'thing':445 'think':42,159 'thorough':669,676 'time':557 'tool':164,347 'topic-agent-skills' 'topic-ai-agents' 'topic-beads' 'topic-claude-code' 'topic-codex' 'topic-cursor' 'topic-developer-tools' 'topic-gemini-cli' 'topic-opencode' 'topic-plugin' 'topic-slash-commands' 'topic-spec-driven-development' 'trace':136 'tracer':349 'track':29,36,45,202,703,710 'tri':86 'uncertainti':298,301 'understand':88 'unexplor':407 'updat':241,245,253,424,518 'url':591 'use':2,188,592,690 'vagu':92,95 'valid':509 'var':580 'verifi':299,515 'vs':587 'way':240 'weak':119,276 'within':171 'without':197,423,475,490 'workflow':77,173,695 'wrong':237,340,444 'yet':265","prices":[{"id":"92e3029d-6be0-4941-ae61-f4b1023a4daf","listingId":"e55f068b-4f06-4549-b59b-4e1698e6ea1d","amountUsd":"0","unit":"free","nativeCurrency":null,"nativeAmount":null,"chain":null,"payTo":null,"paymentMethod":"skill-free","isPrimary":true,"details":{"org":"cofin","category":"flow","install_from":"skills.sh"},"createdAt":"2026-04-23T13:03:58.645Z"}],"sources":[{"listingId":"e55f068b-4f06-4549-b59b-4e1698e6ea1d","source":"github","sourceId":"cofin/flow/deepthink","sourceUrl":"https://github.com/cofin/flow/tree/main/skills/deepthink","isPrimary":false,"firstSeenAt":"2026-04-23T13:03:58.645Z","lastSeenAt":"2026-05-18T19:07:36.078Z"}],"details":{"listingId":"e55f068b-4f06-4549-b59b-4e1698e6ea1d","quickStartSnippet":null,"exampleRequest":null,"exampleResponse":null,"schema":null,"openapiUrl":null,"agentsTxtUrl":null,"citations":[],"useCases":[],"bestFor":[],"notFor":[],"kindDetails":{"org":"cofin","slug":"deepthink","github":{"repo":"cofin/flow","stars":11,"topics":["agent-skills","ai-agents","beads","claude-code","codex","context-driven-development","cursor","developer-tools","gemini-cli","opencode","plugin","slash-commands","spec-driven-development","subagents","tdd","workflow"],"license":"apache-2.0","html_url":"https://github.com/cofin/flow","pushed_at":"2026-04-27T19:07:26Z","description":"Context-Driven Development toolkit for AI agents — spec-first planning, TDD workflow, and Beads integration.","skill_md_sha":"2b2fcb3eac554a204a2168c3d5df15d72fdb3eac","skill_md_path":"skills/deepthink/SKILL.md","default_branch":"main","skill_tree_url":"https://github.com/cofin/flow/tree/main/skills/deepthink"},"layout":"multi","source":"github","category":"flow","frontmatter":{"name":"deepthink","description":"Use when a problem resists quick answers, debugging stalls, analysis feels shallow, confidence is low, hypotheses are competing, reasoning loops repeat, or a hard problem needs evidence tracking."},"skills_sh_url":"https://skills.sh/cofin/flow/deepthink"},"updatedAt":"2026-05-18T19:07:36.078Z"}}