[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7ADE53TAT126XH49BHVKKB6":3,"skill-same:01M7ADE53TAT126XH49BHVKKB6":55},{"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":27,"commitSha":28,"skillDirectory":29,"forksCount":30,"repositoryHtmlUrl":31,"repositoryHomepageUrl":32,"ownerHtmlUrl":33,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":34,"install":35,"resources":48},"01M7ADE53TAT126XH49BHVKKB6","context-hygiene",null,"Use when context is growing large or the agent forgets earlier instructions, when deciding clear vs compact, when processing documents\u002Flogs >50k tokens, when monitoring token cost or limiting subagent fan-out, when choosing a reasoning effort level, or when reducing token\u002Fcompute\u002FMCP costs. Covers lost-in-the-middle, context folding (RLM), cost guards, and difficulty-matched effort.","Leostruka\u002Fdevin-bundle","Leostruka","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F81370685?v=4","01M7AD8MAK63VDNAWRDRRRW04N","leandro ostruka","01M7AD8MBS65YDS1KMMV7YPM07",0,"MIT","https:\u002F\u002Fgithub.com\u002FLeostruka\u002Fdevin-bundle\u002Fblob\u002Fmain\u002Fskills\u002Fcontext-hygiene\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: context-hygiene\ndescription: Use when context is growing large or the agent forgets earlier instructions, when deciding clear vs compact, when processing documents\u002Flogs >50k tokens, when monitoring token cost or limiting subagent fan-out, when choosing a reasoning effort level, or when reducing token\u002Fcompute\u002FMCP costs. Covers lost-in-the-middle, context folding (RLM), cost guards, and difficulty-matched effort.\ntriggers: [user, model]\n---\n\n# Context Hygiene\n\nContext window = input + output tokens the model sees at once. Every system\nprompt, rules file, MCP tool definition, message, and tool result counts\nagainst a hard provider-set limit. Output quality degrades past ~100k active\ntokens (\"dumb zone\") regardless of total window — lost-in-the-middle means\nmiddle detail has weak impact (arXiv:2307.03172). Bigger window ≠ better\nretrieval.\n\n**Default stance:** with SWE-2 (262K) most sessions fit — use context\ndirectly. Apply the techniques below only when actually needed.\n\n## Clear vs Compact\n\n| Action | What it does | When |\n|---|---|---|\n| `clear` | Wipes history | **Default.** Task done or unrelated to prior chat. |\n| `compact` | Summarizes history | Same task continues; losing thread would hurt. Lossy — preserves intent, not facts. |\n\nCompaction leaves \"sediment\" — treat it as escape hatch, not hygiene default.\nIf dense access to early detail is needed, fold (below) instead.\n\n## Prune vs Fold vs Compact vs Summarize\n\n| Action | Mechanism | When |\n|---|---|---|\n| Prune | Per-item keep\u002Ftruncate\u002Fdrop via `extensions\u002Flaya-compactor` (laya `compact-item-v1`, suggestion\u002Fabstain only) | Tool noise fills the window; facts must survive verbatim. |\n| Fold | Offload artifacts to files; peek\u002Fgrep\u002Fsub-query | Dense access to early detail needed; >50k-token artifacts. |\n| Compact | `compact` summary | Escape hatch when pruning cannot reach target size. |\n| Summarize | Manual tl;dr | Almost never inside a session; for handoffs write a doc instead. |\n\nPruning decision policy lives here: pin system + first message + last user\ntask + newest `preserve_recent` messages; abstain means keep (fail-safe);\ndrop applies pair-wise so no `tool_result` is orphaned; survivors stay\nbyte-identical. Escalate to `compact` summary only when pruning cannot\nreach `compact_to`.\n\n**Smart Window params** (adaptive effective window, not fixed recency):\n\n| Param | Value | Basis |\n|---|---|---|\n| `compact_at` | 0.70 x window (SWE-2 262k -> ~183k) | headroom for next tool batch |\n| `compact_to` | 0.40 x window (~105k) | hysteresis, avoids thrash |\n| `preserve_recent` | 6 messages + first message | upstream port default |\n| `keep_threshold` | 0.5, calibrate on real transcripts | laya confidence is entropy-based |\n| `truncate_head` | 300 chars | upstream port default |\n| `max_state` | ~25k est. tokens | fits laya context |\n| `reduction_floor` | \u003C0.25 -> abort, keep original | avoids pointless churn |\n\nCompacting EARLY beats late: the last ~30% of the window is where quality degrades (context pollution), so the threshold fires before the zone, not inside it.\n\n`assist` mode requires `activation_errors` empty plus a calibrated\nthreshold; run shadow first, measure, then promote.\n\n## Lean Context Rules\n\n1. Clear chats between unrelated tasks.\n2. Keep rules files small — they load into *every* conversation. Modularize\n   into skills; see `writing-skills`.\n3. Be paranoid about MCP servers — each injects every tool definition into\n   the system prompt. Audit with `mcp-governance` before adding; keep \u003C10-15\n   tools\u002Fserver.\n4. Don't paste huge documents — write to file, then `read` offset\u002Flimit or\n   `grep`.\n5. Prefer subagents for parallel exploration — each has its own window; only\n   synthesis returns (50-100x savings).\n6. Watch the budget — `context-budget.py` (SessionStart hook).\n7. Guard against overflow — unbounded loops are a cost and correctness risk.\n\n## Context Folding (RLM-style, for >50k-token artifacts)\n\nAdapted from Recursive Language Models (arXiv:2512.24601): treat context as a\nvariable in an environment — peek, grep, partition, sub-query — instead of\nreading it all.\n\nWorkflow: **offload** to file → **peek** (read offset\u002Flimit) → **grep** to\nlocate → **partition** (`split -l 500`) → **sub-query** chunks with\n`researcher` subagents → **verify returns** (Rule 12 spot-checks) →\n**synthesize**.\n\n**Depth rule (critical):** depth=1 only. Subagents must NOT spawn subagents —\ndepth=2 causes overthinking, 95x slower (arXiv:2603.02615).\n\nDon't fold preemptively, don't dump whole files into context, don't use for\nsimple retrieval (grep is cheaper).\n\n**Reversible folds** (`scripts\u002Ffold_store.py`, inspired by billion-context):\nwhen you fold a range the session may need again, persist it with an id —\nthe summary stays in context, the content survives on disk.\n\n```bash\npython skills\u002Fcontext-hygiene\u002Fscripts\u002Ffold_store.py save \u003Cid> \\\n    --summary \"one line\" --content-file huge_log.txt   # fold away\nfold_store.py unfold \u003Cid>      # restore full text when needed\nfold_store.py grep \u003Ckeyword>   # find which fold holds a detail\nfold_store.py list             # ids, summaries, est. tokens\n```\n\nFold ids are durable across sessions (`.devin\u002Fscratch\u002Ffolds\u002F`); delete\nstale folds at task end to keep the store small.\n\n## Cost Guard\n\n| Limit | Action |\n|---|---|\n| 1-3 subagents | Default max; >3 requires explicit user approval (enforced by `validate-tool-args.py`). |\n| Loops\u002Fretries | Bound every loop; each iteration multiplies cost. |\n| Token maxing | Detect effort disproportionate to task difficulty. |\n\n## Output modes (token-saver)\n\nPreset style-of-output fragments injected by `token-saver.py` only when the\nmode changes (state in `.devin\u002Foutput-mode.json`). Set with\n`{\"mode\": \"off|caveman|lean|quiet\"}` in that file; combos like `lean+quiet`\ninject both fragments.\n\n| Mode | Effect |\n|---|---|\n| `off` | Default; no preset (switching back emits one restore notice) |\n| `caveman` | Ultra-telegraphic, bullets only, no preamble |\n| `lean` | Minimal diffs, zero prose around code |\n| `quiet` | No tool-call narration; compact command preference |\n\nRelated: `git-diff-stat.py` adds a one-line `diff-stat:` footer after exec\ncalls when the active mode's preset sets `flags.diff_footer` (caveman, lean);\n`rl-resume-detect.py` writes a `.devin\u002Frts\u002Fintents\u002F` resume intent when a\nStop fires on a rate-limit signature (opt-in, never auto-resumed).\n\n## Cost Optimization Protocol\n\n1. Measure current cost (tokens\u002Frequest, model tier, cache hit rate).\n2. Audit context usage (`mcp-governance` for MCP tool-definition cost).\n3. Right-size effort (see below); paid models only when explicitly approved.\n4. Add caching; shorten prompts; prefer file snippets over full reads.\n5. Re-measure; report tokens\u002Fcost before and after with commands used.\n\n## Code-formatting minification — scoped caution\n\nStripping non-semantic whitespace cuts **input** tokens ~24.5% (arXiv:2508.13666,\nICSE'26 — Java 33.7% avg; tool: Deblank). Verified but **narrow**: measured on\nFIM completion only. Do NOT minify code in agentic\u002Fediting flows —\n\n- SWE-bench agents: input −42% but resolution 50%→38% (arXiv:2606.01326).\n- Whitespace-preserving rewrites flip predictions up to 60% (TokDrift,\n  arXiv:2510.14972).\n- Python: indentation is syntax → ~6.5% gain; minified files break\n  edit-matching, linters, and review.\n- Output stays formatted regardless (−2.9%); \"emit unformatted\" prompting\n  broke Gemini 67%→11% and is model-fragile.\n\nValid as a per-project rule only when ALL hold: model reads but never edits\nthe code (analysis\u002Fcompletion\u002FRAG input); brace-delimited language (not\nPython\u002FYAML); a formatter restores output (Deblank pattern — prettier\u002FCI\nround-trip, disk unchanged); payload is one-shot, not an editing session.\n\n## Effort Calibration\n\nEffort is scarce — match to task difficulty. Maps to `data\u002Fbundle-models.json`:\n\n| Level | model_uid | When |\n|---|---|---|\n| Medium | `swe-2-medium` | Simple tasks, spot fixes, isolated scripts, mechanical refactors. Single test\u002Fbuild verifies. |\n| High | `swe-2-high` | Multi-file tasks with clear dependencies. **Default.** |\n| Max | `swe-2-max` | Open-ended, long-horizon, architecture, novel debugging, or failure at lower effort. |\n\nRules:\n1. **Improve the spec before raising effort** — information quality\n   substitutes for reasoning budget (arXiv:2608.01347).\n2. **Start minimal, expand on failure** (E3: Estimate, Execute, Expand —\n   arXiv:2607.13034; 85% cost cut at equal success on simple edits).\n3. **Never start at Max** for a task untried at Medium. But don't starve hard\n   tasks — higher effort lifted first-try success 28%→89% on a hard build\n   (arXiv:2607.02436).\n4. **Prompt wording allocates work** — \"multiple approaches\" multiplies\n   reasoning 2.4-7.4× with no success gain; \"max certainty\" creates\n   verification loops 18× median cost. State WHAT + acceptance criteria.\n","# Context Hygiene\n\nContext window = input + output tokens the model sees at once. Every system\nprompt, rules file, MCP tool definition, message, and tool result counts\nagainst a hard provider-set limit. Output quality degrades past ~100k active\ntokens (\"dumb zone\") regardless of total window — lost-in-the-middle means\nmiddle detail has weak impact (arXiv:2307.03172). Bigger window ≠ better\nretrieval.\n\n**Default stance:** with SWE-2 (262K) most sessions fit — use context\ndirectly. Apply the techniques below only when actually needed.\n\n## Clear vs Compact\n\n| Action | What it does | When |\n|---|---|---|\n| `clear` | Wipes history | **Default.** Task done or unrelated to prior chat. |\n| `compact` | Summarizes history | Same task continues; losing thread would hurt. Lossy — preserves intent, not facts. |\n\nCompaction leaves \"sediment\" — treat it as escape hatch, not hygiene default.\nIf dense access to early detail is needed, fold (below) instead.\n\n## Prune vs Fold vs Compact vs Summarize\n\n| Action | Mechanism | When |\n|---|---|---|\n| Prune | Per-item keep\u002Ftruncate\u002Fdrop via `extensions\u002Flaya-compactor` (laya `compact-item-v1`, suggestion\u002Fabstain only) | Tool noise fills the window; facts must survive verbatim. |\n| Fold | Offload artifacts to files; peek\u002Fgrep\u002Fsub-query | Dense access to early detail needed; >50k-token artifacts. |\n| Compact | `compact` summary | Escape hatch when pruning cannot reach target size. |\n| Summarize | Manual tl;dr | Almost never inside a session; for handoffs write a doc instead. |\n\nPruning decision policy lives here: pin system + first message + last user\ntask + newest `preserve_recent` messages; abstain means keep (fail-safe);\ndrop applies pair-wise so no `tool_result` is orphaned; survivors stay\nbyte-identical. Escalate to `compact` summary only when pruning cannot\nreach `compact_to`.\n\n**Smart Window params** (adaptive effective window, not fixed recency):\n\n| Param | Value | Basis |\n|---|---|---|\n| `compact_at` | 0.70 x window (SWE-2 262k -> ~183k) | headroom for next tool batch |\n| `compact_to` | 0.40 x window (~105k) | hysteresis, avoids thrash |\n| `preserve_recent` | 6 messages + first message | upstream port default |\n| `keep_threshold` | 0.5, calibrate on real transcripts | laya confidence is entropy-based |\n| `truncate_head` | 300 chars | upstream port default |\n| `max_state` | ~25k est. tokens | fits laya context |\n| `reduction_floor` | \u003C0.25 -> abort, keep original | avoids pointless churn |\n\nCompacting EARLY beats late: the last ~30% of the window is where quality degrades (context pollution), so the threshold fires before the zone, not inside it.\n\n`assist` mode requires `activation_errors` empty plus a calibrated\nthreshold; run shadow first, measure, then promote.\n\n## Lean Context Rules\n\n1. Clear chats between unrelated tasks.\n2. Keep rules files small — they load into *every* conversation. Modularize\n   into skills; see `writing-skills`.\n3. Be paranoid about MCP servers — each injects every tool definition into\n   the system prompt. Audit with `mcp-governance` before adding; keep \u003C10-15\n   tools\u002Fserver.\n4. Don't paste huge documents — write to file, then `read` offset\u002Flimit or\n   `grep`.\n5. Prefer subagents for parallel exploration — each has its own window; only\n   synthesis returns (50-100x savings).\n6. Watch the budget — `context-budget.py` (SessionStart hook).\n7. Guard against overflow — unbounded loops are a cost and correctness risk.\n\n## Context Folding (RLM-style, for >50k-token artifacts)\n\nAdapted from Recursive Language Models (arXiv:2512.24601): treat context as a\nvariable in an environment — peek, grep, partition, sub-query — instead of\nreading it all.\n\nWorkflow: **offload** to file → **peek** (read offset\u002Flimit) → **grep** to\nlocate → **partition** (`split -l 500`) → **sub-query** chunks with\n`researcher` subagents → **verify returns** (Rule 12 spot-checks) →\n**synthesize**.\n\n**Depth rule (critical):** depth=1 only. Subagents must NOT spawn subagents —\ndepth=2 causes overthinking, 95x slower (arXiv:2603.02615).\n\nDon't fold preemptively, don't dump whole files into context, don't use for\nsimple retrieval (grep is cheaper).\n\n**Reversible folds** (`scripts\u002Ffold_store.py`, inspired by billion-context):\nwhen you fold a range the session may need again, persist it with an id —\nthe summary stays in context, the content survives on disk.\n\n```bash\npython skills\u002Fcontext-hygiene\u002Fscripts\u002Ffold_store.py save \u003Cid> \\\n    --summary \"one line\" --content-file huge_log.txt   # fold away\nfold_store.py unfold \u003Cid>      # restore full text when needed\nfold_store.py grep \u003Ckeyword>   # find which fold holds a detail\nfold_store.py list             # ids, summaries, est. tokens\n```\n\nFold ids are durable across sessions (`.devin\u002Fscratch\u002Ffolds\u002F`); delete\nstale folds at task end to keep the store small.\n\n## Cost Guard\n\n| Limit | Action |\n|---|---|\n| 1-3 subagents | Default max; >3 requires explicit user approval (enforced by `validate-tool-args.py`). |\n| Loops\u002Fretries | Bound every loop; each iteration multiplies cost. |\n| Token maxing | Detect effort disproportionate to task difficulty. |\n\n## Output modes (token-saver)\n\nPreset style-of-output fragments injected by `token-saver.py` only when the\nmode changes (state in `.devin\u002Foutput-mode.json`). Set with\n`{\"mode\": \"off|caveman|lean|quiet\"}` in that file; combos like `lean+quiet`\ninject both fragments.\n\n| Mode | Effect |\n|---|---|\n| `off` | Default; no preset (switching back emits one restore notice) |\n| `caveman` | Ultra-telegraphic, bullets only, no preamble |\n| `lean` | Minimal diffs, zero prose around code |\n| `quiet` | No tool-call narration; compact command preference |\n\nRelated: `git-diff-stat.py` adds a one-line `diff-stat:` footer after exec\ncalls when the active mode's preset sets `flags.diff_footer` (caveman, lean);\n`rl-resume-detect.py` writes a `.devin\u002Frts\u002Fintents\u002F` resume intent when a\nStop fires on a rate-limit signature (opt-in, never auto-resumed).\n\n## Cost Optimization Protocol\n\n1. Measure current cost (tokens\u002Frequest, model tier, cache hit rate).\n2. Audit context usage (`mcp-governance` for MCP tool-definition cost).\n3. Right-size effort (see below); paid models only when explicitly approved.\n4. Add caching; shorten prompts; prefer file snippets over full reads.\n5. Re-measure; report tokens\u002Fcost before and after with commands used.\n\n## Code-formatting minification — scoped caution\n\nStripping non-semantic whitespace cuts **input** tokens ~24.5% (arXiv:2508.13666,\nICSE'26 — Java 33.7% avg; tool: Deblank). Verified but **narrow**: measured on\nFIM completion only. Do NOT minify code in agentic\u002Fediting flows —\n\n- SWE-bench agents: input −42% but resolution 50%→38% (arXiv:2606.01326).\n- Whitespace-preserving rewrites flip predictions up to 60% (TokDrift,\n  arXiv:2510.14972).\n- Python: indentation is syntax → ~6.5% gain; minified files break\n  edit-matching, linters, and review.\n- Output stays formatted regardless (−2.9%); \"emit unformatted\" prompting\n  broke Gemini 67%→11% and is model-fragile.\n\nValid as a per-project rule only when ALL hold: model reads but never edits\nthe code (analysis\u002Fcompletion\u002FRAG input); brace-delimited language (not\nPython\u002FYAML); a formatter restores output (Deblank pattern — prettier\u002FCI\nround-trip, disk unchanged); payload is one-shot, not an editing session.\n\n## Effort Calibration\n\nEffort is scarce — match to task difficulty. Maps to `data\u002Fbundle-models.json`:\n\n| Level | model_uid | When |\n|---|---|---|\n| Medium | `swe-2-medium` | Simple tasks, spot fixes, isolated scripts, mechanical refactors. Single test\u002Fbuild verifies. |\n| High | `swe-2-high` | Multi-file tasks with clear dependencies. **Default.** |\n| Max | `swe-2-max` | Open-ended, long-horizon, architecture, novel debugging, or failure at lower effort. |\n\nRules:\n1. **Improve the spec before raising effort** — information quality\n   substitutes for reasoning budget (arXiv:2608.01347).\n2. **Start minimal, expand on failure** (E3: Estimate, Execute, Expand —\n   arXiv:2607.13034; 85% cost cut at equal success on simple edits).\n3. **Never start at Max** for a task untried at Medium. But don't starve hard\n   tasks — higher effort lifted first-try success 28%→89% on a hard build\n   (arXiv:2607.02436).\n4. **Prompt wording allocates work** — \"multiple approaches\" multiplies\n   reasoning 2.4-7.4× with no success gain; \"max certainty\" creates\n   verification loops 18× median cost. State WHAT + acceptance criteria.\n",{"name":5,"triggers":24,"description":7},[25,26],"user","model","main","02b95ddb62959a00b3c31442c2046aca0d19563c","skills\u002Fcontext-hygiene",1,"https:\u002F\u002Fgithub.com\u002FLeostruka\u002Fdevin-bundle","https:\u002F\u002Fleostruka.github.io\u002Fdevin-bundle\u002F","https:\u002F\u002Fgithub.com\u002FLeostruka","2026-11-15 05:19:58",{"source":8,"skillName":5,"commands":36,"prompt":46,"zipAvailable":47},[37,40,43],{"tool":38,"command":39},"npx","npx skills add Leostruka\u002Fdevin-bundle --skill context-hygiene",{"tool":41,"command":42},"bunx","bunx skills add Leostruka\u002Fdevin-bundle --skill context-hygiene",{"tool":44,"command":45},"pnpm","pnpm dlx skills add Leostruka\u002Fdevin-bundle --skill context-hygiene","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FLeostruka\u002Fdevin-bundle\u002Fblob\u002Fmain\u002Fskills\u002Fcontext-hygiene\u002FSKILL.md",true,[49],{"path":50,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":53,"sourceHtmlUrl":54},"scripts\u002Ffold_store.py","SCRIPT","脚本",4302,"https:\u002F\u002Fgithub.com\u002FLeostruka\u002Fdevin-bundle\u002Fblob\u002Fmain\u002Fskills\u002Fcontext-hygiene\u002Fscripts\u002Ffold_store.py",[56,63,70,77,84,91],{"skillNo":57,"name":58,"nameZh":6,"description":59,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":60,"categories":61,"keywords":62,"publishedAt":19},"01M7E0K3MG2XC4FSRV0MK6SFR1","montador-de-ideias","Use when assembling, expanding, or elevating a raw project idea or initial prompt into an ultra-detailed, first-principles dossier and compiled master prompt with journalistic OSINT market research, Tech Lead and Staff Architect justifications, atomic and modular hardening\u002Foptimization, and high-impact contact surfaces.","https:\u002F\u002Fgithub.com\u002FLeostruka\u002Fdevin-bundle\u002Fblob\u002Fmain\u002Fskills\u002Fmontador-de-ideias\u002FSKILL.md",[],[],{"skillNo":64,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":67,"categories":68,"keywords":69,"publishedAt":19},"01M7E0J7JJWZ8QMK01RZ3B3C9Z","git-github-adaptive-engineering","Use when any task touches git history, GitHub issues\u002FPRs\u002Freleases\u002FCI, repo governance, or branch and merge decisions. Runs the Fase Zero situational-awareness battery, picks one of five delivery flows via the adaptive decision engine, and enforces commit\u002FPR information-density standards. Orchestrates the gh, git-workflows, github-actions, pr, and debug\u002Ffinish skills.","https:\u002F\u002Fgithub.com\u002FLeostruka\u002Fdevin-bundle\u002Fblob\u002Fmain\u002Fskills\u002Fgit-github-adaptive-engineering\u002FSKILL.md",[],[],{"skillNo":71,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":19},"01M7E0GJFRPFEVQQMXEM58EWDM","adaptive-agenda-calendar-routine-engine","Use when planning or executing work across hours, days, weeks, or sprints. Turns the calendar into a strict time-blocked agenda (HH:MM-HH:MM blocks carrying a routine profile, expected inputs, and a handoff output artifact), runs the 5-stage cognitive flow (Radar and Hunt -> Triage plus Steve Jobs conception -> workbench manifest -> >=32-cycle loop -> Close plus Open-Minded Scrum), and unfolds macro calendar milestones into micro agenda blocks through extensions\u002Fgcal\u002Fwrapper.py.","https:\u002F\u002Fgithub.com\u002FLeostruka\u002Fdevin-bundle\u002Fblob\u002Fmain\u002Fskills\u002Fadaptive-agenda-calendar-routine-engine\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":80,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":19},"01M7AE2K9FTY0BVS48N0A9JFKF","demo","Use when demonstrating a skill in a fixture project.","https:\u002F\u002Fgithub.com\u002FLeostruka\u002Fdevin-bundle\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fdevin-config\u002Fproject\u002F.devin\u002Fskills\u002Fdemo\u002FSKILL.md",[],[],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":19},"01M7AE2G3DP38PTV3B1N7MW5TS","duplicate-skill","Use when testing duplicate skill name detection.","https:\u002F\u002Fgithub.com\u002FLeostruka\u002Fdevin-bundle\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fdevin-config\u002Fproject\u002F.devin\u002Fskills\u002Fbeta\u002FSKILL.md",[],[],{"skillNo":92,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":93,"categories":94,"keywords":95,"publishedAt":19},"01M7AE2DBR8ZHT6T875MJ392GK","https:\u002F\u002Fgithub.com\u002FLeostruka\u002Fdevin-bundle\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fdevin-config\u002Fproject\u002F.devin\u002Fskills\u002Falpha\u002FSKILL.md",[],[]]