[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7F7DHC51FPWAVZBXC3RC05R":3,"skill-same:01M7F7DHC51FPWAVZBXC3RC05R":46},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":6,"ownerHtmlUrl":29,"ownerBlogUrl":6,"ownerDescription":30,"updatedAt":31,"install":32,"resources":45},"01M7F7DHC51FPWAVZBXC3RC05R","interactive-problem-solving",null,"Solve or debug information-acquisition tasks where queries or evaluator feedback must reduce uncertainty until a hidden state, answer, or contract is sufficiently determined. Use for interactive problems, output-only tasks with a checker, and custom-judge workflows when the central difficulty is the protocol contract, query design, hypothesis elimination, query budget, transcript behavior, noise, or adversarial feedback. Do not use for offline output-only or scorer-only optimization whose feedback occurs only between development runs; use heuristic search. Use reactive-online-decision-problem-solving when reward-bearing actions receive live feedback that changes later decisions.","lza6\u002FArmor-piercing-system-prompts","lza6","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F128880206?v=4","01M7EB3K7WPX5T28EJ2NVHE6EG","Chinese-tingfeng","01M7EB3K97E1KQ16QWKJMFNGX8",11,"MIT","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F06-前端与设计\u002Falgorithmic-problem-solving\u002Fsub-skills\u002Finteractive-problem-solving\u002FSKILL.md",[],[],"2026-11-16 10:05:34",false,"---\nname: interactive-problem-solving\ndescription: >-\n  Solve or debug information-acquisition tasks where queries or evaluator\n  feedback must reduce uncertainty until a hidden state, answer, or contract is\n  sufficiently determined. Use for interactive problems, output-only tasks\n  with a checker, and custom-judge workflows when the central difficulty is the\n  protocol contract, query design, hypothesis elimination, query budget,\n  transcript behavior, noise, or adversarial feedback. Do not use for offline\n  output-only or scorer-only optimization whose feedback occurs only between\n  development runs; use heuristic search. Use reactive-online-decision-problem-solving when\n  reward-bearing actions receive live feedback that changes later decisions.\n---\n\n# Interactive Problem Solving\n\nReduce uncertainty until the answer is uniquely or sufficiently determined:\n\n```\nunderstand the interaction contract\n        -> design a query\n        -> observe and update hypotheses\n        -> recover the hidden state or final answer\n```\n\nThe central question is: **What should be asked next?** The main value of an action is the information it reveals.\n\n## Boundary with decision-making\n\nUse this skill when termination means that the answer or relevant hidden state is known well enough. Use [reactive online decision problem solving](..\u002Freactive-online-decision-problem-solving\u002FSKILL.md) when termination is a round horizon, task completion, or budget exhaustion and actions must earn immediate or future reward while controlling risk.\n\nFor an output-only task with a checker or a custom judge, treat an evaluated candidate as a query only when its verdict, score, or diagnostic is primarily being used to infer hidden acceptance conditions, state, or evaluator behavior. If the contract is already known and the goal is offline repeated score improvement, route to [heuristic search](..\u002F..\u002Freferences\u002Fheuristic-search.md). Route to [Reactive](..\u002Freactive-online-decision-problem-solving\u002FSKILL.md) only when reward-bearing actions receive live feedback that changes later decisions.\n\n## 1. Establish the literal contract\n\nRead the statement and every supplied interactor, tester, checker, scorer, runner, and configuration file relevant to the information channel. Record:\n\n```\ninitial judge output and ordering\nlegal query\u002Fcandidate syntax and parameter bounds\nexact deterministic response function or stochastic observation model\ncounted queries\u002Fevaluations\u002Factions and total budget\nrequired flush points and process wiring\nspecial replies: error, -1, success, EOF, timeout\nfinal-answer syntax and whether it consumes budget\nstate reset or persistence across cases\u002Frounds\u002Fsubmissions\ncorrectness and score dependence on answers, queries, actions, and time\n```\n\nReplay sequential updates, rounding, accumulators, and state transitions literally. When prose and executable behavior differ, document both, derive their legal intersection, and use the intersection when it preserves correctness and budget. If a stronger solution depends on executable-only behavior, name and directly test that dependency.\n\nOnly when a concrete unresolved deterministic validity or score question requires independent reconstruction, use [checker and local evaluation](..\u002Fchecker-and-local-evaluation\u002FSKILL.md) rather than treating checker calls as interactive queries. Coherent accepted official evidence does not trigger that route by itself.\n\n## 2. Model hypotheses and identifiability\n\nFormalize deterministic interaction as:\n\n```\nH_t       hidden states consistent with the transcript through time t\nR(h, q)   exact response to query q in hidden state h\nH_{t+1} = {h in H_t : R(h, q_t) matches the observed response}\n```\n\nFor stochastic feedback, maintain likelihoods, posterior weights, or confidence sets rather than eliminating a hypothesis after one mismatch.\n\nBefore coding a strategy, test identifiability:\n\n- Enumerate tiny hidden-state spaces and response partitions.\n- Search for symmetric or globally indistinguishable states.\n- Partition the hidden states into `K` final-answer classes so one same sufficient accepted answer works for every state in a class.\n- If every response has at most `b > 1` outcomes, compare the budget with the necessary lower bound `ceil(log_b K)`; use `K = |H|` only when exact hidden-state recovery is required, and account for unbalanced partitions and indistinguishability.\n- Prove that the planned transcript separates every pair of classes that require different final answers, not merely sampled states.\n\nExact state recovery is unnecessary when every remaining hypothesis implies the same accepted answer. State that equivalence explicitly.\n\n## 3. Choose the next query\n\nDerive candidate queries from the exact response semantics. Rank them by the criterion matching the judge:\n\n- minimize the worst-case remaining hypothesis count;\n- maximize justified expected entropy or uncertainty reduction;\n- distinguish a targeted pair or equivalence class;\n- exploit group testing, coding, parity, algebraic cancellation, balanced separators, or batched independent features;\n- retain separation margin under noise or adversarial perturbation;\n- include query cost when queries have unequal price.\n\nFor an adversarial but consistent judge, optimize the worst response partition and preserve the complete feasible set or an equivalent invariant. For known stochastic noise, use likelihood updates, repeated measurements, sequential tests, robust estimators, or error-correcting separation. For unknown noise, reserve calibration queries and avoid brittle hard elimination.\n\nDo not multiply per-query success probabilities without establishing independence or a valid conditional bound.\n\n## 4. Isolate protocol mechanics\n\nKeep the solver's information logic behind a small interface:\n\n```\nread_initial()\nask(query) -> response\nupdate(response)\nanswer(solution)\n```\n\n`ask` owns serialization, budget accounting, flushing, reply validation, and immediate termination on judge error or EOF. Keep diagnostics off stdout. Put buffering, deadline, numeric, and target-toolchain fixes in [contest solver engineering](..\u002Fcontest-solver-engineering\u002FSKILL.md); do not mix them into the query-selection proof.\n\n## 5. Validate with a simulator when diagnostic\n\nRoute to [checker and local evaluation](..\u002Fchecker-and-local-evaluation\u002FSKILL.md) when a concrete protocol symptom, disputed state transition, unexplained official response, or strategy hypothesis needs a process-level simulator\u002Finteractor, real-pipe runner, transcript contract, or independent state\u002Freward replay. If accepted official interactions have stable protocol behavior and coherent feedback, do not build a simulator solely because this section was reached; defer it until it can expose or discriminate a problem. Once routed, use [validation and experiments](..\u002Fvalidation-and-experiments\u002FSKILL.md#7-interactive-and-reactive-evaluator-validation) to incorporate those artifacts into the broader validation ladder and solver comparison loop.\n\nWhen evaluator work is routed and the repository uses `testlib.h`, first keep the evaluator contract and independence rules in Checker and Local Evaluation, then follow [testlib C++ judging](..\u002Ftestlib-cpp-judging\u002FSKILL.md) for the concrete interactor\u002Fchecker API and build commands.\n\nWhen such a simulator or transcript campaign is justified, validate the information strategy itself with:\n\n- exhaustive tiny hidden states;\n- exact budget exhaustion and final-answer accounting;\n- a worst-case consistent response chooser;\n- fixed-seed noisy transcripts when applicable;\n- collision search over final remaining hypothesis classes;\n- early success, error, EOF, and malformed-reply paths.\n\nThe solver must never observe simulator-only truth.\n\n## 6. Safety and termination\n\n- Check the budget before emitting each query or candidate evaluation.\n- Keep a legal final answer for every reachable belief state when partial credit or early termination exists.\n- Reserve enough budget and time to serialize the answer and complete mandatory protocol steps.\n- Use deterministic tie-breaking while debugging.\n- Terminate as soon as the remaining hypotheses imply one sufficient answer; do not spend queries merely to identify irrelevant hidden details.\n\n## Diagnostics\n\n| Symptom | Test first |\n|---|---|\n| Immediate protocol failure | Initial read order, syntax, indexing, flush, special reply |\n| Correct tiny cases but query blow-up | Partition balance, symmetry, repeated information |\n| Decoder collision | Literal response semantics and global hypothesis enumeration |\n| Works in functions but hangs remotely | Real bidirectional pipes, buffering, EOF, timeout |\n| Noise causes unstable elimination | Likelihood model, dependence, calibration, robust separation |\n| Nonzero feedback but non-perfect result | Whether feedback measures correctness, query count, or quality |\n\nReturn the recovered contract, query invariant\u002Fcriterion, identifiability argument, implementation, simulator\u002Ftranscript commands actually run, maximum observed query count, and remaining assumptions.\n","# Interactive Problem Solving\n\nReduce uncertainty until the answer is uniquely or sufficiently determined:\n\n```\nunderstand the interaction contract\n        -> design a query\n        -> observe and update hypotheses\n        -> recover the hidden state or final answer\n```\n\nThe central question is: **What should be asked next?** The main value of an action is the information it reveals.\n\n## Boundary with decision-making\n\nUse this skill when termination means that the answer or relevant hidden state is known well enough. Use [reactive online decision problem solving](..\u002Freactive-online-decision-problem-solving\u002FSKILL.md) when termination is a round horizon, task completion, or budget exhaustion and actions must earn immediate or future reward while controlling risk.\n\nFor an output-only task with a checker or a custom judge, treat an evaluated candidate as a query only when its verdict, score, or diagnostic is primarily being used to infer hidden acceptance conditions, state, or evaluator behavior. If the contract is already known and the goal is offline repeated score improvement, route to [heuristic search](..\u002F..\u002Freferences\u002Fheuristic-search.md). Route to [Reactive](..\u002Freactive-online-decision-problem-solving\u002FSKILL.md) only when reward-bearing actions receive live feedback that changes later decisions.\n\n## 1. Establish the literal contract\n\nRead the statement and every supplied interactor, tester, checker, scorer, runner, and configuration file relevant to the information channel. Record:\n\n```\ninitial judge output and ordering\nlegal query\u002Fcandidate syntax and parameter bounds\nexact deterministic response function or stochastic observation model\ncounted queries\u002Fevaluations\u002Factions and total budget\nrequired flush points and process wiring\nspecial replies: error, -1, success, EOF, timeout\nfinal-answer syntax and whether it consumes budget\nstate reset or persistence across cases\u002Frounds\u002Fsubmissions\ncorrectness and score dependence on answers, queries, actions, and time\n```\n\nReplay sequential updates, rounding, accumulators, and state transitions literally. When prose and executable behavior differ, document both, derive their legal intersection, and use the intersection when it preserves correctness and budget. If a stronger solution depends on executable-only behavior, name and directly test that dependency.\n\nOnly when a concrete unresolved deterministic validity or score question requires independent reconstruction, use [checker and local evaluation](..\u002Fchecker-and-local-evaluation\u002FSKILL.md) rather than treating checker calls as interactive queries. Coherent accepted official evidence does not trigger that route by itself.\n\n## 2. Model hypotheses and identifiability\n\nFormalize deterministic interaction as:\n\n```\nH_t       hidden states consistent with the transcript through time t\nR(h, q)   exact response to query q in hidden state h\nH_{t+1} = {h in H_t : R(h, q_t) matches the observed response}\n```\n\nFor stochastic feedback, maintain likelihoods, posterior weights, or confidence sets rather than eliminating a hypothesis after one mismatch.\n\nBefore coding a strategy, test identifiability:\n\n- Enumerate tiny hidden-state spaces and response partitions.\n- Search for symmetric or globally indistinguishable states.\n- Partition the hidden states into `K` final-answer classes so one same sufficient accepted answer works for every state in a class.\n- If every response has at most `b > 1` outcomes, compare the budget with the necessary lower bound `ceil(log_b K)`; use `K = |H|` only when exact hidden-state recovery is required, and account for unbalanced partitions and indistinguishability.\n- Prove that the planned transcript separates every pair of classes that require different final answers, not merely sampled states.\n\nExact state recovery is unnecessary when every remaining hypothesis implies the same accepted answer. State that equivalence explicitly.\n\n## 3. Choose the next query\n\nDerive candidate queries from the exact response semantics. Rank them by the criterion matching the judge:\n\n- minimize the worst-case remaining hypothesis count;\n- maximize justified expected entropy or uncertainty reduction;\n- distinguish a targeted pair or equivalence class;\n- exploit group testing, coding, parity, algebraic cancellation, balanced separators, or batched independent features;\n- retain separation margin under noise or adversarial perturbation;\n- include query cost when queries have unequal price.\n\nFor an adversarial but consistent judge, optimize the worst response partition and preserve the complete feasible set or an equivalent invariant. For known stochastic noise, use likelihood updates, repeated measurements, sequential tests, robust estimators, or error-correcting separation. For unknown noise, reserve calibration queries and avoid brittle hard elimination.\n\nDo not multiply per-query success probabilities without establishing independence or a valid conditional bound.\n\n## 4. Isolate protocol mechanics\n\nKeep the solver's information logic behind a small interface:\n\n```\nread_initial()\nask(query) -> response\nupdate(response)\nanswer(solution)\n```\n\n`ask` owns serialization, budget accounting, flushing, reply validation, and immediate termination on judge error or EOF. Keep diagnostics off stdout. Put buffering, deadline, numeric, and target-toolchain fixes in [contest solver engineering](..\u002Fcontest-solver-engineering\u002FSKILL.md); do not mix them into the query-selection proof.\n\n## 5. Validate with a simulator when diagnostic\n\nRoute to [checker and local evaluation](..\u002Fchecker-and-local-evaluation\u002FSKILL.md) when a concrete protocol symptom, disputed state transition, unexplained official response, or strategy hypothesis needs a process-level simulator\u002Finteractor, real-pipe runner, transcript contract, or independent state\u002Freward replay. If accepted official interactions have stable protocol behavior and coherent feedback, do not build a simulator solely because this section was reached; defer it until it can expose or discriminate a problem. Once routed, use [validation and experiments](..\u002Fvalidation-and-experiments\u002FSKILL.md#7-interactive-and-reactive-evaluator-validation) to incorporate those artifacts into the broader validation ladder and solver comparison loop.\n\nWhen evaluator work is routed and the repository uses `testlib.h`, first keep the evaluator contract and independence rules in Checker and Local Evaluation, then follow [testlib C++ judging](..\u002Ftestlib-cpp-judging\u002FSKILL.md) for the concrete interactor\u002Fchecker API and build commands.\n\nWhen such a simulator or transcript campaign is justified, validate the information strategy itself with:\n\n- exhaustive tiny hidden states;\n- exact budget exhaustion and final-answer accounting;\n- a worst-case consistent response chooser;\n- fixed-seed noisy transcripts when applicable;\n- collision search over final remaining hypothesis classes;\n- early success, error, EOF, and malformed-reply paths.\n\nThe solver must never observe simulator-only truth.\n\n## 6. Safety and termination\n\n- Check the budget before emitting each query or candidate evaluation.\n- Keep a legal final answer for every reachable belief state when partial credit or early termination exists.\n- Reserve enough budget and time to serialize the answer and complete mandatory protocol steps.\n- Use deterministic tie-breaking while debugging.\n- Terminate as soon as the remaining hypotheses imply one sufficient answer; do not spend queries merely to identify irrelevant hidden details.\n\n## Diagnostics\n\n| Symptom | Test first |\n|---|---|\n| Immediate protocol failure | Initial read order, syntax, indexing, flush, special reply |\n| Correct tiny cases but query blow-up | Partition balance, symmetry, repeated information |\n| Decoder collision | Literal response semantics and global hypothesis enumeration |\n| Works in functions but hangs remotely | Real bidirectional pipes, buffering, EOF, timeout |\n| Noise causes unstable elimination | Likelihood model, dependence, calibration, robust separation |\n| Nonzero feedback but non-perfect result | Whether feedback measures correctness, query count, or quality |\n\nReturn the recovered contract, query invariant\u002Fcriterion, identifiability argument, implementation, simulator\u002Ftranscript commands actually run, maximum observed query count, and remaining assumptions.\n",{"name":5,"description":7},"main","d69096e239f2a89aabc2cd078c47664a9d3b1e11","06-前端与设计\u002Falgorithmic-problem-solving\u002Fsub-skills\u002Finteractive-problem-solving",3,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts","https:\u002F\u002Fgithub.com\u002Flza6","个人vx：Tf00798\r\n个人✈：@TFGY999","2026-10-08 14:29:22",{"source":8,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add lza6\u002FArmor-piercing-system-prompts --skill interactive-problem-solving",{"tool":38,"command":39},"bunx","bunx skills add lza6\u002FArmor-piercing-system-prompts --skill interactive-problem-solving",{"tool":41,"command":42},"pnpm","pnpm dlx skills add lza6\u002FArmor-piercing-system-prompts --skill interactive-problem-solving","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F06-前端与设计\u002Falgorithmic-problem-solving\u002Fsub-skills\u002Finteractive-problem-solving\u002FSKILL.md",true,[],[47,54,59,66,73,80],{"skillNo":48,"name":49,"nameZh":6,"description":50,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":51,"categories":52,"keywords":53,"publishedAt":19},"01M7FFG485KY83TD6HSC4YPCMC","obsidian-clipper-template-creator","Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobsidian-clipper-template-creator\u002FSKILL.md",[],[],{"skillNo":55,"name":49,"nameZh":6,"description":50,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":56,"categories":57,"keywords":58,"publishedAt":19},"01M7FFG0T6AZ0STEJPSB3CX84K","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobsidian-clipper-template-creator-antigravity-awesome-skills-main\u002FSKILL.md",[],[],{"skillNo":60,"name":61,"nameZh":6,"description":62,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":63,"categories":64,"keywords":65,"publishedAt":19},"01M7FFFY0QEM14SKWPDYGMXM2P","python-api","Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP\u002FmAP semantics, and source-style import diagnostics.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002Fsub-skills\u002Fpython-api\u002FSKILL.md",[],[],{"skillNo":67,"name":68,"nameZh":6,"description":69,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":70,"categories":71,"keywords":72,"publishedAt":19},"01M7FFFVZQZ2GJ0C0TGP7JHD20","file-evaluation","Folder-based PASCAL VOC AP\u002FmAP evaluation from ground-truth and detection text files.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002Fsub-skills\u002Ffile-evaluation\u002FSKILL.md",[],[],{"skillNo":74,"name":75,"nameZh":6,"description":76,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":77,"categories":78,"keywords":79,"publishedAt":19},"01M7FFFSFNYSE77SJ6PE1CK2DA","object-detection-metrics","Guides legacy Object-Detection-Metrics PASCAL VOC AP\u002FmAP evaluation from detection text folders and direct Python API classes.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002FSKILL.md",[],[],{"skillNo":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":84,"categories":85,"keywords":86,"publishedAt":19},"01M7FFFP0GCK1RVXBCPQ8APBMT","nx-workspace-patterns","Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fnx-workspace-patterns\u002FSKILL.md",[],[]]