{ "schemaVersion": "1.0", "item": { "slug": "agent-development", "name": "Agent Development", "source": "tencent", "type": "skill", "category": "AI 智能", "sourceUrl": "https://clawhub.ai/Veeramanikandanr48/agent-development", "canonicalUrl": "https://clawhub.ai/Veeramanikandanr48/agent-development", "targetPlatform": "OpenClaw" }, "install": { "downloadMode": "redirect", "downloadUrl": "/downloads/agent-development", "sourceDownloadUrl": "https://wry-manatee-359.convex.site/api/v1/download?slug=agent-development", "sourcePlatform": "tencent", "targetPlatform": "OpenClaw", "installMethod": "Manual import", "extraction": "Extract archive", "prerequisites": [ "OpenClaw" ], "packageFormat": "ZIP package", "includedAssets": [ ".claude-plugin/plugin.json", "README.md", "SKILL.md", "rules/agent-memory-limits.md", "rules/agent-pattern.md", "rules/agent-self-documentation.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. Then review README.md for any prerequisites, environment setup, or post-install checks. 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. Then review README.md for any prerequisites, environment setup, or post-install checks. Summarize what changed and any follow-up checks I should run." } ] }, "sourceHealth": { "source": "tencent", "status": "healthy", "reason": "direct_download_ok", "recommendedAction": "download", "checkedAt": "2026-04-23T16:43:11.935Z", "expiresAt": "2026-04-30T16:43:11.935Z", "httpStatus": 200, "finalUrl": "https://wry-manatee-359.convex.site/api/v1/download?slug=4claw-imageboard", "contentType": "application/zip", "probeMethod": "head", "details": { "probeUrl": "https://wry-manatee-359.convex.site/api/v1/download?slug=4claw-imageboard", "contentDisposition": "attachment; filename=\"4claw-imageboard-1.0.1.zip\"", "redirectLocation": null, "bodySnippet": null }, "scope": "source", "summary": "Source download looks usable.", "detail": "Yavira can redirect you to the upstream package for this source.", "primaryActionLabel": "Download for OpenClaw", "primaryActionHref": "/downloads/agent-development" }, "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/agent-development", "agentPageUrl": "https://openagent3.xyz/skills/agent-development/agent", "manifestUrl": "https://openagent3.xyz/skills/agent-development/agent.json", "briefUrl": "https://openagent3.xyz/skills/agent-development/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. Then review README.md for any prerequisites, environment setup, or post-install checks. 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. Then review README.md for any prerequisites, environment setup, or post-install checks. Summarize what changed and any follow-up checks I should run." } ] }, "documentation": { "source": "clawhub", "primaryDoc": "SKILL.md", "sections": [ { "title": "Agent Development for Claude Code", "body": "Build effective custom agents for Claude Code with proper delegation, tool access, and prompt design." }, { "title": "Agent Description Pattern", "body": "The description field determines whether Claude will automatically delegate tasks." }, { "title": "Strong Trigger Pattern", "body": "---\nname: agent-name\ndescription: |\n [Role] specialist. MUST BE USED when [specific triggers].\n Use PROACTIVELY for [task category].\n Keywords: [trigger words]\ntools: Read, Write, Edit, Glob, Grep, Bash\nmodel: sonnet\n---" }, { "title": "Weak vs Strong Descriptions", "body": "Weak (won't auto-delegate)Strong (auto-delegates)\"Analyzes screenshots for issues\"\"Visual QA specialist. MUST BE USED when analyzing screenshots. Use PROACTIVELY for visual QA.\"\"Runs Playwright scripts\"\"Playwright specialist. MUST BE USED when running Playwright scripts. Use PROACTIVELY for browser automation.\"\n\nKey phrases:\n\n\"MUST BE USED when...\"\n\"Use PROACTIVELY for...\"\nInclude trigger keywords" }, { "title": "Delegation Mechanisms", "body": "Explicit: Task tool subagent_type: \"agent-name\" - always works\nAutomatic: Claude matches task to agent description - requires strong phrasing\n\nSession restart required after creating/modifying agents." }, { "title": "Tool Access Principle", "body": "If an agent doesn't need Bash, don't give it Bash.\n\nAgent needs to...Give toolsDon't giveCreate files onlyRead, Write, Edit, Glob, GrepBashRun scripts/CLIsRead, Write, Edit, Glob, Grep, Bash—Read/audit onlyRead, Glob, GrepWrite, Edit, Bash\n\nWhy? Models default to cat > file << 'EOF' heredocs instead of Write tool. Each bash command requires approval, causing dozens of prompts per agent run." }, { "title": "Allowlist Pattern", "body": "Instead of restricting Bash, allowlist safe commands in .claude/settings.json:\n\n{\n \"permissions\": {\n \"allow\": [\n \"Write\", \"Edit\", \"WebFetch(domain:*)\",\n \"Bash(cd *)\", \"Bash(cp *)\", \"Bash(mkdir *)\", \"Bash(ls *)\",\n \"Bash(cat *)\", \"Bash(head *)\", \"Bash(tail *)\", \"Bash(grep *)\",\n \"Bash(diff *)\", \"Bash(mv *)\", \"Bash(touch *)\", \"Bash(file *)\"\n ]\n }\n}" }, { "title": "Model Selection (Quality First)", "body": "Don't downgrade quality to work around issues - fix root causes instead.\n\nModelUse ForOpusCreative work (page building, design, content) - quality mattersSonnetMost agents - content, code, research (default)HaikuOnly script runners where quality doesn't matter" }, { "title": "Root Cause Fix (REQUIRED)", "body": "Add to ~/.bashrc or ~/.zshrc:\n\nexport NODE_OPTIONS=\"--max-old-space-size=16384\"\n\nIncreases Node.js heap from 4GB to 16GB." }, { "title": "Parallel Limits (Even With Fix)", "body": "Agent TypeMax ParallelNotesAny agents2-3Context accumulates; batch then pauseHeavy creative (Opus)1-2Uses more memory" }, { "title": "Recovery", "body": "source ~/.bashrc or restart terminal\nNODE_OPTIONS=\"--max-old-space-size=16384\" claude\nCheck what files exist, continue from there" }, { "title": "Sub-Agent vs Remote API", "body": "Always prefer Task sub-agents over remote API calls.\n\nAspectRemote API CallTask Sub-AgentTool accessNoneFull (Read, Grep, Write, Bash)File readingMust pass all content in promptCan read files iterativelyCross-referencingSingle context windowCan reason across documentsDecision qualityGeneric suggestionsSpecific decisions with rationaleOutput quality~100 lines typical600+ lines with specifics\n\n// ❌ WRONG - Remote API call\nconst response = await fetch('https://api.anthropic.com/v1/messages', {...})\n\n// ✅ CORRECT - Use Task tool\n// Invoke Task with subagent_type: \"general-purpose\"" }, { "title": "Declarative Over Imperative", "body": "Describe what to accomplish, not how to use tools." }, { "title": "Wrong (Imperative)", "body": "### Check for placeholders\n```bash\ngrep -r \"PLACEHOLDER:\" build/*.html\n\n### Right (Declarative)\n\n```markdown\n### Check for placeholders\nSearch all HTML files in build/ for:\n- PLACEHOLDER: comments\n- TODO or TBD markers\n- Template brackets like [Client Name]\n\nAny match = incomplete content." }, { "title": "What to Include", "body": "IncludeSkipTask goal and contextExplicit bash/tool commandsInput file paths\"Use X tool to...\"Output file paths and formatStep-by-step tool invocationsSuccess/failure criteriaShell pipeline syntaxBlocking checks (prerequisites)Micromanaged workflowsQuality checklists" }, { "title": "Self-Documentation Principle", "body": "\"Agents that won't have your context must be able to reproduce the behaviour independently.\"\n\nEvery improvement must be encoded into the agent's prompt, not left as implicit knowledge." }, { "title": "What to Encode", "body": "DiscoveryWhere to CaptureBug fix patternAgent's \"Corrections\" or \"Common Issues\" sectionQuality requirementAgent's \"Quality Checklist\" sectionFile path conventionAgent's \"Output\" sectionTool usage patternAgent's \"Process\" sectionBlocking prerequisiteAgent's \"Blocking Check\" section" }, { "title": "Test: Would a Fresh Agent Succeed?", "body": "Before completing any agent improvement:\n\nRead the agent prompt as if you have no context\nAsk: Could a new session follow this and produce the same quality?\nIf no: Add missing instructions, patterns, or references" }, { "title": "Anti-Patterns", "body": "Anti-PatternWhy It Fails\"As we discussed earlier...\"No prior context existsRelying on files read during devAgent may not read same filesAssuming knowledge from errorsAgent won't see your debugging\"Just like the home page\"Agent hasn't built home page" }, { "title": "Agent Prompt Structure", "body": "Effective agent prompts include:\n\n## Your Role\n[What the agent does]\n\n## Blocking Check\n[Prerequisites that must exist]\n\n## Input\n[What files to read]\n\n## Process\n[Step-by-step with encoded learnings]\n\n## Output\n[Exact file paths and formats]\n\n## Quality Checklist\n[Verification steps including learned gotchas]\n\n## Common Issues\n[Patterns discovered during development]" }, { "title": "Pipeline Agents", "body": "When inserting a new agent into a numbered pipeline (e.g., HTML-01 → HTML-05 → HTML-11):\n\nMust UpdateWhatNew agent\"Workflow Position\" diagram + \"Next\" fieldPredecessor agentIts \"Next\" field to point to new agent\n\nCommon bug: New agent is \"orphaned\" because predecessor still points to old next agent.\n\nVerification:\n\ngrep -n \"Next:.*→\\|Then.*runs next\" .claude/agents/*.md" }, { "title": "The Sweet Spot", "body": "Best use case: Tasks that are repetitive but require judgment.\n\nExample: Auditing 70 skills manually = tedious. But each audit needs intelligence (check docs, compare versions, decide what to fix). Perfect for parallel agents with clear instructions.\n\nNot good for:\n\nSimple tasks (just do them)\nHighly creative tasks (need human direction)\nTasks requiring cross-file coordination (agents work independently)" }, { "title": "Effective Prompt Template", "body": "For each [item]:\n1. Read [source file]\n2. Verify with [external check - npm view, API call, etc.]\n3. Check [authoritative source]\n4. Score/evaluate\n5. FIX issues found ← Critical instruction\n\nKey elements:\n\n\"FIX issues found\" - Without this, agents only report. With it, they take action.\nExact file paths - Prevents ambiguity\nOutput format template - Ensures consistent, parseable reports\nBatch size ~5 items - Enough work to be efficient, not so much that failures cascade" }, { "title": "Workflow Pattern", "body": "1. ME: Launch 2-3 parallel agents with identical prompt, different item lists\n2. AGENTS: Work in parallel (read → verify → check → edit → report)\n3. AGENTS: Return structured reports (score, status, fixes applied, files modified)\n4. ME: Review changes (git status, spot-check diffs)\n5. ME: Commit in batches with meaningful changelog\n6. ME: Push and update progress tracking\n\nWhy agents don't commit: Allows human review, batching, and clean commit history." }, { "title": "Signs a Task Fits This Pattern", "body": "Good fit:\n\nSame steps repeated for many items\nEach item requires judgment (not just transformation)\nItems are independent (no cross-item dependencies)\nClear success criteria (score, pass/fail, etc.)\nAuthoritative source exists to verify against\n\nBad fit:\n\nItems depend on each other's results\nRequires creative/subjective decisions\nSingle complex task (use regular agent instead)\nNeeds human input mid-process" }, { "title": "Agent Frontmatter Template", "body": "---\nname: my-agent\ndescription: |\n [Role] specialist. MUST BE USED when [triggers].\n Use PROACTIVELY for [task category].\n Keywords: [trigger words]\ntools: Read, Write, Edit, Glob, Grep, Bash\nmodel: sonnet\n---" }, { "title": "Fix Bash Approval Spam", "body": "Remove Bash from tools if not needed\nPut critical instructions FIRST (right after frontmatter)\nUse allowlists in .claude/settings.json" }, { "title": "Memory Crash Recovery", "body": "export NODE_OPTIONS=\"--max-old-space-size=16384\"\nsource ~/.bashrc && claude" } ], "body": "Agent Development for Claude Code\n\nBuild effective custom agents for Claude Code with proper delegation, tool access, and prompt design.\n\nAgent Description Pattern\n\nThe description field determines whether Claude will automatically delegate tasks.\n\nStrong Trigger Pattern\n---\nname: agent-name\ndescription: |\n [Role] specialist. MUST BE USED when [specific triggers].\n Use PROACTIVELY for [task category].\n Keywords: [trigger words]\ntools: Read, Write, Edit, Glob, Grep, Bash\nmodel: sonnet\n---\n\nWeak vs Strong Descriptions\nWeak (won't auto-delegate)\tStrong (auto-delegates)\n\"Analyzes screenshots for issues\"\t\"Visual QA specialist. MUST BE USED when analyzing screenshots. Use PROACTIVELY for visual QA.\"\n\"Runs Playwright scripts\"\t\"Playwright specialist. MUST BE USED when running Playwright scripts. Use PROACTIVELY for browser automation.\"\n\nKey phrases:\n\n\"MUST BE USED when...\"\n\"Use PROACTIVELY for...\"\nInclude trigger keywords\nDelegation Mechanisms\nExplicit: Task tool subagent_type: \"agent-name\" - always works\nAutomatic: Claude matches task to agent description - requires strong phrasing\n\nSession restart required after creating/modifying agents.\n\nTool Access Principle\n\nIf an agent doesn't need Bash, don't give it Bash.\n\nAgent needs to...\tGive tools\tDon't give\nCreate files only\tRead, Write, Edit, Glob, Grep\tBash\nRun scripts/CLIs\tRead, Write, Edit, Glob, Grep, Bash\t—\nRead/audit only\tRead, Glob, Grep\tWrite, Edit, Bash\n\nWhy? Models default to cat > file << 'EOF' heredocs instead of Write tool. Each bash command requires approval, causing dozens of prompts per agent run.\n\nAllowlist Pattern\n\nInstead of restricting Bash, allowlist safe commands in .claude/settings.json:\n\n{\n \"permissions\": {\n \"allow\": [\n \"Write\", \"Edit\", \"WebFetch(domain:*)\",\n \"Bash(cd *)\", \"Bash(cp *)\", \"Bash(mkdir *)\", \"Bash(ls *)\",\n \"Bash(cat *)\", \"Bash(head *)\", \"Bash(tail *)\", \"Bash(grep *)\",\n \"Bash(diff *)\", \"Bash(mv *)\", \"Bash(touch *)\", \"Bash(file *)\"\n ]\n }\n}\n\nModel Selection (Quality First)\n\nDon't downgrade quality to work around issues - fix root causes instead.\n\nModel\tUse For\nOpus\tCreative work (page building, design, content) - quality matters\nSonnet\tMost agents - content, code, research (default)\nHaiku\tOnly script runners where quality doesn't matter\nMemory Limits\nRoot Cause Fix (REQUIRED)\n\nAdd to ~/.bashrc or ~/.zshrc:\n\nexport NODE_OPTIONS=\"--max-old-space-size=16384\"\n\n\nIncreases Node.js heap from 4GB to 16GB.\n\nParallel Limits (Even With Fix)\nAgent Type\tMax Parallel\tNotes\nAny agents\t2-3\tContext accumulates; batch then pause\nHeavy creative (Opus)\t1-2\tUses more memory\nRecovery\nsource ~/.bashrc or restart terminal\nNODE_OPTIONS=\"--max-old-space-size=16384\" claude\nCheck what files exist, continue from there\nSub-Agent vs Remote API\n\nAlways prefer Task sub-agents over remote API calls.\n\nAspect\tRemote API Call\tTask Sub-Agent\nTool access\tNone\tFull (Read, Grep, Write, Bash)\nFile reading\tMust pass all content in prompt\tCan read files iteratively\nCross-referencing\tSingle context window\tCan reason across documents\nDecision quality\tGeneric suggestions\tSpecific decisions with rationale\nOutput quality\t~100 lines typical\t600+ lines with specifics\n// ❌ WRONG - Remote API call\nconst response = await fetch('https://api.anthropic.com/v1/messages', {...})\n\n// ✅ CORRECT - Use Task tool\n// Invoke Task with subagent_type: \"general-purpose\"\n\nDeclarative Over Imperative\n\nDescribe what to accomplish, not how to use tools.\n\nWrong (Imperative)\n### Check for placeholders\n```bash\ngrep -r \"PLACEHOLDER:\" build/*.html\n\n\n### Right (Declarative)\n\n```markdown\n### Check for placeholders\nSearch all HTML files in build/ for:\n- PLACEHOLDER: comments\n- TODO or TBD markers\n- Template brackets like [Client Name]\n\nAny match = incomplete content.\n\nWhat to Include\nInclude\tSkip\nTask goal and context\tExplicit bash/tool commands\nInput file paths\t\"Use X tool to...\"\nOutput file paths and format\tStep-by-step tool invocations\nSuccess/failure criteria\tShell pipeline syntax\nBlocking checks (prerequisites)\tMicromanaged workflows\nQuality checklists\t\nSelf-Documentation Principle\n\n\"Agents that won't have your context must be able to reproduce the behaviour independently.\"\n\nEvery improvement must be encoded into the agent's prompt, not left as implicit knowledge.\n\nWhat to Encode\nDiscovery\tWhere to Capture\nBug fix pattern\tAgent's \"Corrections\" or \"Common Issues\" section\nQuality requirement\tAgent's \"Quality Checklist\" section\nFile path convention\tAgent's \"Output\" section\nTool usage pattern\tAgent's \"Process\" section\nBlocking prerequisite\tAgent's \"Blocking Check\" section\nTest: Would a Fresh Agent Succeed?\n\nBefore completing any agent improvement:\n\nRead the agent prompt as if you have no context\nAsk: Could a new session follow this and produce the same quality?\nIf no: Add missing instructions, patterns, or references\nAnti-Patterns\nAnti-Pattern\tWhy It Fails\n\"As we discussed earlier...\"\tNo prior context exists\nRelying on files read during dev\tAgent may not read same files\nAssuming knowledge from errors\tAgent won't see your debugging\n\"Just like the home page\"\tAgent hasn't built home page\nAgent Prompt Structure\n\nEffective agent prompts include:\n\n## Your Role\n[What the agent does]\n\n## Blocking Check\n[Prerequisites that must exist]\n\n## Input\n[What files to read]\n\n## Process\n[Step-by-step with encoded learnings]\n\n## Output\n[Exact file paths and formats]\n\n## Quality Checklist\n[Verification steps including learned gotchas]\n\n## Common Issues\n[Patterns discovered during development]\n\nPipeline Agents\n\nWhen inserting a new agent into a numbered pipeline (e.g., HTML-01 → HTML-05 → HTML-11):\n\nMust Update\tWhat\nNew agent\t\"Workflow Position\" diagram + \"Next\" field\nPredecessor agent\tIts \"Next\" field to point to new agent\n\nCommon bug: New agent is \"orphaned\" because predecessor still points to old next agent.\n\nVerification:\n\ngrep -n \"Next:.*→\\|Then.*runs next\" .claude/agents/*.md\n\nThe Sweet Spot\n\nBest use case: Tasks that are repetitive but require judgment.\n\nExample: Auditing 70 skills manually = tedious. But each audit needs intelligence (check docs, compare versions, decide what to fix). Perfect for parallel agents with clear instructions.\n\nNot good for:\n\nSimple tasks (just do them)\nHighly creative tasks (need human direction)\nTasks requiring cross-file coordination (agents work independently)\nEffective Prompt Template\nFor each [item]:\n1. Read [source file]\n2. Verify with [external check - npm view, API call, etc.]\n3. Check [authoritative source]\n4. Score/evaluate\n5. FIX issues found ← Critical instruction\n\n\nKey elements:\n\n\"FIX issues found\" - Without this, agents only report. With it, they take action.\nExact file paths - Prevents ambiguity\nOutput format template - Ensures consistent, parseable reports\nBatch size ~5 items - Enough work to be efficient, not so much that failures cascade\nWorkflow Pattern\n1. ME: Launch 2-3 parallel agents with identical prompt, different item lists\n2. AGENTS: Work in parallel (read → verify → check → edit → report)\n3. AGENTS: Return structured reports (score, status, fixes applied, files modified)\n4. ME: Review changes (git status, spot-check diffs)\n5. ME: Commit in batches with meaningful changelog\n6. ME: Push and update progress tracking\n\n\nWhy agents don't commit: Allows human review, batching, and clean commit history.\n\nSigns a Task Fits This Pattern\n\nGood fit:\n\nSame steps repeated for many items\nEach item requires judgment (not just transformation)\nItems are independent (no cross-item dependencies)\nClear success criteria (score, pass/fail, etc.)\nAuthoritative source exists to verify against\n\nBad fit:\n\nItems depend on each other's results\nRequires creative/subjective decisions\nSingle complex task (use regular agent instead)\nNeeds human input mid-process\nQuick Reference\nAgent Frontmatter Template\n---\nname: my-agent\ndescription: |\n [Role] specialist. MUST BE USED when [triggers].\n Use PROACTIVELY for [task category].\n Keywords: [trigger words]\ntools: Read, Write, Edit, Glob, Grep, Bash\nmodel: sonnet\n---\n\nFix Bash Approval Spam\nRemove Bash from tools if not needed\nPut critical instructions FIRST (right after frontmatter)\nUse allowlists in .claude/settings.json\nMemory Crash Recovery\nexport NODE_OPTIONS=\"--max-old-space-size=16384\"\nsource ~/.bashrc && claude" }, "trust": { "sourceLabel": "tencent", "provenanceUrl": "https://clawhub.ai/Veeramanikandanr48/agent-development", "publisherUrl": "https://clawhub.ai/Veeramanikandanr48/agent-development", "owner": "Veeramanikandanr48", "version": "0.1.0", "license": null, "verificationStatus": "Indexed source record" }, "links": { "detailUrl": "https://openagent3.xyz/skills/agent-development", "downloadUrl": "https://openagent3.xyz/downloads/agent-development", "agentUrl": "https://openagent3.xyz/skills/agent-development/agent", "manifestUrl": "https://openagent3.xyz/skills/agent-development/agent.json", "briefUrl": "https://openagent3.xyz/skills/agent-development/agent.md" } }