{ "schemaVersion": "1.0", "item": { "slug": "self-improving", "name": "Self-Improving Agent (With Self-Reflection)", "source": "tencent", "type": "skill", "category": "AI 智能", "sourceUrl": "https://clawhub.ai/ivangdavila/self-improving", "canonicalUrl": "https://clawhub.ai/ivangdavila/self-improving", "targetPlatform": "OpenClaw" }, "install": { "downloadMode": "redirect", "downloadUrl": "/downloads/self-improving", "sourceDownloadUrl": "https://wry-manatee-359.convex.site/api/v1/download?slug=self-improving", "sourcePlatform": "tencent", "targetPlatform": "OpenClaw", "installMethod": "Manual import", "extraction": "Extract archive", "prerequisites": [ "OpenClaw" ], "packageFormat": "ZIP package", "includedAssets": [ "HEARTBEAT.md", "SKILL.md", "boundaries.md", "corrections.md", "heartbeat-rules.md", "heartbeat-state.md" ], "primaryDoc": "SKILL.md", "quickSetup": [ "Download the package from Yavira.", "Extract the archive and review SKILL.md first.", "Import or place the package into your OpenClaw setup." ], "agentAssist": { "summary": "Hand the extracted package to your coding agent with a concrete install brief instead of figuring it out manually.", "steps": [ "Download the package from Yavira.", "Extract it into a folder your agent can access.", "Paste one of the prompts below and point your agent at the extracted folder." ], "prompts": [ { "label": "New install", "body": "I downloaded a skill package from Yavira. Read SKILL.md from the extracted folder and install it by following the included instructions. Tell me what you changed and call out any manual steps you could not complete." }, { "label": "Upgrade existing", "body": "I downloaded an updated skill package from Yavira. Read SKILL.md from the extracted folder, compare it with my current installation, and upgrade it while preserving any custom configuration unless the package docs explicitly say otherwise. Summarize what changed and any follow-up checks I should run." } ] }, "sourceHealth": { "source": "tencent", "slug": "self-improving", "status": "healthy", "reason": "direct_download_ok", "recommendedAction": "download", "checkedAt": "2026-05-03T15:21:19.746Z", "expiresAt": "2026-05-10T15:21:19.746Z", "httpStatus": 200, "finalUrl": "https://wry-manatee-359.convex.site/api/v1/download?slug=self-improving", "contentType": "application/zip", "probeMethod": "head", "details": { "probeUrl": "https://wry-manatee-359.convex.site/api/v1/download?slug=self-improving", "contentDisposition": "attachment; filename=\"self-improving-1.2.16.zip\"", "redirectLocation": null, "bodySnippet": null, "slug": "self-improving" }, "scope": "item", "summary": "Item download looks usable.", "detail": "Yavira can redirect you to the upstream package for this item.", "primaryActionLabel": "Download for OpenClaw", "primaryActionHref": "/downloads/self-improving" }, "validation": { "installChecklist": [ "Use the Yavira download entry.", "Review SKILL.md after the package is downloaded.", "Confirm the extracted package contains the expected setup assets." ], "postInstallChecks": [ "Confirm the extracted package includes the expected docs or setup files.", "Validate the skill or prompts are available in your target agent workspace.", "Capture any manual follow-up steps the agent could not complete." ] }, "downloadPageUrl": "https://openagent3.xyz/downloads/self-improving", "agentPageUrl": "https://openagent3.xyz/skills/self-improving/agent", "manifestUrl": "https://openagent3.xyz/skills/self-improving/agent.json", "briefUrl": "https://openagent3.xyz/skills/self-improving/agent.md" }, "agentAssist": { "summary": "Hand the extracted package to your coding agent with a concrete install brief instead of figuring it out manually.", "steps": [ "Download the package from Yavira.", "Extract it into a folder your agent can access.", "Paste one of the prompts below and point your agent at the extracted folder." ], "prompts": [ { "label": "New install", "body": "I downloaded a skill package from Yavira. Read SKILL.md from the extracted folder and install it by following the included instructions. Tell me what you changed and call out any manual steps you could not complete." }, { "label": "Upgrade existing", "body": "I downloaded an updated skill package from Yavira. Read SKILL.md from the extracted folder, compare it with my current installation, and upgrade it while preserving any custom configuration unless the package docs explicitly say otherwise. Summarize what changed and any follow-up checks I should run." } ] }, "documentation": { "source": "clawhub", "primaryDoc": "SKILL.md", "sections": [ { "title": "When to Use", "body": "User corrects you or points out mistakes. You complete significant work and want to evaluate the outcome. You notice something in your own output that could be better. Knowledge should compound over time without manual maintenance." }, { "title": "Architecture", "body": "Memory lives in ~/self-improving/ with tiered structure. If ~/self-improving/ does not exist, run setup.md.\nWorkspace setup should add the standard self-improving steering to the workspace AGENTS, SOUL, and HEARTBEAT.md files, with recurring maintenance routed through heartbeat-rules.md.\n\n~/self-improving/\n├── memory.md # HOT: ≤100 lines, always loaded\n├── index.md # Topic index with line counts\n├── heartbeat-state.md # Heartbeat state: last run, reviewed change, action notes\n├── projects/ # Per-project learnings\n├── domains/ # Domain-specific (code, writing, comms)\n├── archive/ # COLD: decayed patterns\n└── corrections.md # Last 50 corrections log" }, { "title": "Quick Reference", "body": "TopicFileSetup guidesetup.mdHeartbeat state templateheartbeat-state.mdMemory templatememory-template.mdWorkspace heartbeat snippetHEARTBEAT.mdHeartbeat rulesheartbeat-rules.mdLearning mechanicslearning.mdSecurity boundariesboundaries.mdScaling rulesscaling.mdMemory operationsoperations.mdSelf-reflection logreflections.mdOpenClaw HEARTBEAT seedopenclaw-heartbeat.md" }, { "title": "Requirements", "body": "No credentials required\nNo extra binaries required\nOptional installation of the Proactivity skill may require network access" }, { "title": "Learning Signals", "body": "Log automatically when you notice these patterns:\n\nCorrections → add to corrections.md, evaluate for memory.md:\n\n\"No, that's not right...\"\n\"Actually, it should be...\"\n\"You're wrong about...\"\n\"I prefer X, not Y\"\n\"Remember that I always...\"\n\"I told you before...\"\n\"Stop doing X\"\n\"Why do you keep...\"\n\nPreference signals → add to memory.md if explicit:\n\n\"I like when you...\"\n\"Always do X for me\"\n\"Never do Y\"\n\"My style is...\"\n\"For [project], use...\"\n\nPattern candidates → track, promote after 3x:\n\nSame instruction repeated 3+ times\nWorkflow that works well repeatedly\nUser praises specific approach\n\nIgnore (don't log):\n\nOne-time instructions (\"do X now\")\nContext-specific (\"in this file...\")\nHypotheticals (\"what if...\")" }, { "title": "Self-Reflection", "body": "After completing significant work, pause and evaluate:\n\nDid it meet expectations? — Compare outcome vs intent\nWhat could be better? — Identify improvements for next time\nIs this a pattern? — If yes, log to corrections.md\n\nWhen to self-reflect:\n\nAfter completing a multi-step task\nAfter receiving feedback (positive or negative)\nAfter fixing a bug or mistake\nWhen you notice your output could be better\n\nLog format:\n\nCONTEXT: [type of task]\nREFLECTION: [what I noticed]\nLESSON: [what to do differently]\n\nExample:\n\nCONTEXT: Building Flutter UI\nREFLECTION: Spacing looked off, had to redo\nLESSON: Check visual spacing before showing user\n\nSelf-reflection entries follow the same promotion rules: 3x applied successfully → promote to HOT." }, { "title": "Quick Queries", "body": "User saysAction\"What do you know about X?\"Search all tiers for X\"What have you learned?\"Show last 10 from corrections.md\"Show my patterns\"List memory.md (HOT)\"Show [project] patterns\"Load projects/{name}.md\"What's in warm storage?\"List files in projects/ + domains/\"Memory stats\"Show counts per tier\"Forget X\"Remove from all tiers (confirm first)\"Export memory\"ZIP all files" }, { "title": "Memory Stats", "body": "On \"memory stats\" request, report:\n\n📊 Self-Improving Memory\n\nHOT (always loaded):\n memory.md: X entries\n\nWARM (load on demand):\n projects/: X files\n domains/: X files\n\nCOLD (archived):\n archive/: X files\n\nRecent activity (7 days):\n Corrections logged: X\n Promotions to HOT: X\n Demotions to WARM: X" }, { "title": "Common Traps", "body": "TrapWhy It FailsBetter MoveLearning from silenceCreates false rulesWait for explicit correction or repeated evidencePromoting too fastPollutes HOT memoryKeep new lessons tentative until repeatedReading every namespaceWastes contextLoad only HOT plus the smallest matching filesCompaction by deletionLoses trust and historyMerge, summarize, or demote instead" }, { "title": "1. Learn from Corrections and Self-Reflection", "body": "Log when user explicitly corrects you\nLog when you identify improvements in your own work\nNever infer from silence alone\nAfter 3 identical lessons → ask to confirm as rule" }, { "title": "2. Tiered Storage", "body": "TierLocationSize LimitBehaviorHOTmemory.md≤100 linesAlways loadedWARMprojects/, domains/≤200 lines eachLoad on context matchCOLDarchive/UnlimitedLoad on explicit query" }, { "title": "3. Automatic Promotion/Demotion", "body": "Pattern used 3x in 7 days → promote to HOT\nPattern unused 30 days → demote to WARM\nPattern unused 90 days → archive to COLD\nNever delete without asking" }, { "title": "4. Namespace Isolation", "body": "Project patterns stay in projects/{name}.md\nGlobal preferences in HOT tier (memory.md)\nDomain patterns (code, writing) in domains/\nCross-namespace inheritance: global → domain → project" }, { "title": "5. Conflict Resolution", "body": "When patterns contradict:\n\nMost specific wins (project > domain > global)\nMost recent wins (same level)\nIf ambiguous → ask user" }, { "title": "6. Compaction", "body": "When file exceeds limit:\n\nMerge similar corrections into single rule\nArchive unused patterns\nSummarize verbose entries\nNever lose confirmed preferences" }, { "title": "7. Transparency", "body": "Every action from memory → cite source: \"Using X (from projects/foo.md:12)\"\nWeekly digest available: patterns learned, demoted, archived\nFull export on demand: all files as ZIP" }, { "title": "8. Security Boundaries", "body": "See boundaries.md — never store credentials, health data, third-party info." }, { "title": "9. Graceful Degradation", "body": "If context limit hit:\n\nLoad only memory.md (HOT)\nLoad relevant namespace on demand\nNever fail silently — tell user what's not loaded" }, { "title": "Scope", "body": "This skill ONLY:\n\nLearns from user corrections and self-reflection\nStores preferences in local files (~/self-improving/)\nMaintains heartbeat state in ~/self-improving/heartbeat-state.md when the workspace integrates heartbeat\nReads its own memory files on activation\n\nThis skill NEVER:\n\nAccesses calendar, email, or contacts\nMakes network requests\nReads files outside ~/self-improving/\nInfers preferences from silence or observation\nDeletes or blindly rewrites self-improving memory during heartbeat cleanup\nModifies its own SKILL.md" }, { "title": "Data Storage", "body": "Local state lives in ~/self-improving/:\n\nmemory.md for HOT rules and confirmed preferences\ncorrections.md for explicit corrections and reusable lessons\nprojects/ and domains/ for scoped patterns\narchive/ for decayed or inactive patterns\nheartbeat-state.md for recurring maintenance markers" }, { "title": "Related Skills", "body": "Install with clawhub install if user confirms:\n\nmemory — Long-term memory patterns for agents\nlearning — Adaptive teaching and explanation\ndecide — Auto-learn decision patterns\nescalate — Know when to ask vs act autonomously" }, { "title": "Feedback", "body": "If useful: clawhub star self-improving\nStay updated: clawhub sync" } ], "body": "When to Use\n\nUser corrects you or points out mistakes. You complete significant work and want to evaluate the outcome. You notice something in your own output that could be better. Knowledge should compound over time without manual maintenance.\n\nArchitecture\n\nMemory lives in ~/self-improving/ with tiered structure. If ~/self-improving/ does not exist, run setup.md. Workspace setup should add the standard self-improving steering to the workspace AGENTS, SOUL, and HEARTBEAT.md files, with recurring maintenance routed through heartbeat-rules.md.\n\n~/self-improving/\n├── memory.md # HOT: ≤100 lines, always loaded\n├── index.md # Topic index with line counts\n├── heartbeat-state.md # Heartbeat state: last run, reviewed change, action notes\n├── projects/ # Per-project learnings\n├── domains/ # Domain-specific (code, writing, comms)\n├── archive/ # COLD: decayed patterns\n└── corrections.md # Last 50 corrections log\n\nQuick Reference\nTopic\tFile\nSetup guide\tsetup.md\nHeartbeat state template\theartbeat-state.md\nMemory template\tmemory-template.md\nWorkspace heartbeat snippet\tHEARTBEAT.md\nHeartbeat rules\theartbeat-rules.md\nLearning mechanics\tlearning.md\nSecurity boundaries\tboundaries.md\nScaling rules\tscaling.md\nMemory operations\toperations.md\nSelf-reflection log\treflections.md\nOpenClaw HEARTBEAT seed\topenclaw-heartbeat.md\nRequirements\nNo credentials required\nNo extra binaries required\nOptional installation of the Proactivity skill may require network access\nLearning Signals\n\nLog automatically when you notice these patterns:\n\nCorrections → add to corrections.md, evaluate for memory.md:\n\n\"No, that's not right...\"\n\"Actually, it should be...\"\n\"You're wrong about...\"\n\"I prefer X, not Y\"\n\"Remember that I always...\"\n\"I told you before...\"\n\"Stop doing X\"\n\"Why do you keep...\"\n\nPreference signals → add to memory.md if explicit:\n\n\"I like when you...\"\n\"Always do X for me\"\n\"Never do Y\"\n\"My style is...\"\n\"For [project], use...\"\n\nPattern candidates → track, promote after 3x:\n\nSame instruction repeated 3+ times\nWorkflow that works well repeatedly\nUser praises specific approach\n\nIgnore (don't log):\n\nOne-time instructions (\"do X now\")\nContext-specific (\"in this file...\")\nHypotheticals (\"what if...\")\nSelf-Reflection\n\nAfter completing significant work, pause and evaluate:\n\nDid it meet expectations? — Compare outcome vs intent\nWhat could be better? — Identify improvements for next time\nIs this a pattern? — If yes, log to corrections.md\n\nWhen to self-reflect:\n\nAfter completing a multi-step task\nAfter receiving feedback (positive or negative)\nAfter fixing a bug or mistake\nWhen you notice your output could be better\n\nLog format:\n\nCONTEXT: [type of task]\nREFLECTION: [what I noticed]\nLESSON: [what to do differently]\n\n\nExample:\n\nCONTEXT: Building Flutter UI\nREFLECTION: Spacing looked off, had to redo\nLESSON: Check visual spacing before showing user\n\n\nSelf-reflection entries follow the same promotion rules: 3x applied successfully → promote to HOT.\n\nQuick Queries\nUser says\tAction\n\"What do you know about X?\"\tSearch all tiers for X\n\"What have you learned?\"\tShow last 10 from corrections.md\n\"Show my patterns\"\tList memory.md (HOT)\n\"Show [project] patterns\"\tLoad projects/{name}.md\n\"What's in warm storage?\"\tList files in projects/ + domains/\n\"Memory stats\"\tShow counts per tier\n\"Forget X\"\tRemove from all tiers (confirm first)\n\"Export memory\"\tZIP all files\nMemory Stats\n\nOn \"memory stats\" request, report:\n\n📊 Self-Improving Memory\n\nHOT (always loaded):\n memory.md: X entries\n\nWARM (load on demand):\n projects/: X files\n domains/: X files\n\nCOLD (archived):\n archive/: X files\n\nRecent activity (7 days):\n Corrections logged: X\n Promotions to HOT: X\n Demotions to WARM: X\n\nCommon Traps\nTrap\tWhy It Fails\tBetter Move\nLearning from silence\tCreates false rules\tWait for explicit correction or repeated evidence\nPromoting too fast\tPollutes HOT memory\tKeep new lessons tentative until repeated\nReading every namespace\tWastes context\tLoad only HOT plus the smallest matching files\nCompaction by deletion\tLoses trust and history\tMerge, summarize, or demote instead\nCore Rules\n1. Learn from Corrections and Self-Reflection\nLog when user explicitly corrects you\nLog when you identify improvements in your own work\nNever infer from silence alone\nAfter 3 identical lessons → ask to confirm as rule\n2. Tiered Storage\nTier\tLocation\tSize Limit\tBehavior\nHOT\tmemory.md\t≤100 lines\tAlways loaded\nWARM\tprojects/, domains/\t≤200 lines each\tLoad on context match\nCOLD\tarchive/\tUnlimited\tLoad on explicit query\n3. Automatic Promotion/Demotion\nPattern used 3x in 7 days → promote to HOT\nPattern unused 30 days → demote to WARM\nPattern unused 90 days → archive to COLD\nNever delete without asking\n4. Namespace Isolation\nProject patterns stay in projects/{name}.md\nGlobal preferences in HOT tier (memory.md)\nDomain patterns (code, writing) in domains/\nCross-namespace inheritance: global → domain → project\n5. Conflict Resolution\n\nWhen patterns contradict:\n\nMost specific wins (project > domain > global)\nMost recent wins (same level)\nIf ambiguous → ask user\n6. Compaction\n\nWhen file exceeds limit:\n\nMerge similar corrections into single rule\nArchive unused patterns\nSummarize verbose entries\nNever lose confirmed preferences\n7. Transparency\nEvery action from memory → cite source: \"Using X (from projects/foo.md:12)\"\nWeekly digest available: patterns learned, demoted, archived\nFull export on demand: all files as ZIP\n8. Security Boundaries\n\nSee boundaries.md — never store credentials, health data, third-party info.\n\n9. Graceful Degradation\n\nIf context limit hit:\n\nLoad only memory.md (HOT)\nLoad relevant namespace on demand\nNever fail silently — tell user what's not loaded\nScope\n\nThis skill ONLY:\n\nLearns from user corrections and self-reflection\nStores preferences in local files (~/self-improving/)\nMaintains heartbeat state in ~/self-improving/heartbeat-state.md when the workspace integrates heartbeat\nReads its own memory files on activation\n\nThis skill NEVER:\n\nAccesses calendar, email, or contacts\nMakes network requests\nReads files outside ~/self-improving/\nInfers preferences from silence or observation\nDeletes or blindly rewrites self-improving memory during heartbeat cleanup\nModifies its own SKILL.md\nData Storage\n\nLocal state lives in ~/self-improving/:\n\nmemory.md for HOT rules and confirmed preferences\ncorrections.md for explicit corrections and reusable lessons\nprojects/ and domains/ for scoped patterns\narchive/ for decayed or inactive patterns\nheartbeat-state.md for recurring maintenance markers\nRelated Skills\n\nInstall with clawhub install if user confirms:\n\nmemory — Long-term memory patterns for agents\nlearning — Adaptive teaching and explanation\ndecide — Auto-learn decision patterns\nescalate — Know when to ask vs act autonomously\nFeedback\nIf useful: clawhub star self-improving\nStay updated: clawhub sync" }, "trust": { "sourceLabel": "tencent", "provenanceUrl": "https://clawhub.ai/ivangdavila/self-improving", "publisherUrl": "https://clawhub.ai/ivangdavila/self-improving", "owner": "ivangdavila", "version": "1.2.16", "license": null, "verificationStatus": "Indexed source record" }, "links": { "detailUrl": "https://openagent3.xyz/skills/self-improving", "downloadUrl": "https://openagent3.xyz/downloads/self-improving", "agentUrl": "https://openagent3.xyz/skills/self-improving/agent", "manifestUrl": "https://openagent3.xyz/skills/self-improving/agent.json", "briefUrl": "https://openagent3.xyz/skills/self-improving/agent.md" } }