{"id":"7903177e-877f-467a-892c-47080610ad86","slug":"upgradeables","kind":"skill","name":"upgradeables","summary":"Open, model-agnostic building blocks and copy-ready prompts for AI Skills, reasoning workflows, validators, and agent systems.","owner":"robkazi52","repoUrl":"https://github.com/robkazi52/upgradeables","license":"Apache-2.0","licenseTier":"permissive","trust":"community","stars":0,"match":null,"fewStars":true,"format":"skill","createdAt":"2026-09-03T20:10:43.000Z","addedAt":"2026-10-21T11:33:54.775Z","logoUrl":"https://avatars.githubusercontent.com/u/257547886?v=4","linkOnly":false,"collection":null,"skillPath":null,"categories":["writing"],"tags":["agents","ai","ai-safety","ai-tools","llm","llm-agents","open-source","orchestration"],"runtime":"python","local":true,"cloudCapable":false,"cloudCapability":"unknown","cloudReadyOnPlatform":false,"needsApiKey":false,"reviewed":false,"syncedAt":"2026-10-23T05:07:02.634Z","sample":false,"url":"https://conduits.dev/tools/upgradeables","language":"Python","transports":["stdio"],"install":{"pypi":"upgradeables-registry"},"requirements":{"env":[],"upstreamMayCharge":false},"tools":null,"readme":"# Upgradeables\n\nReusable, model-agnostic building blocks for AI work. Use them directly in a\nchat, combine them into a reusable Skill, or adapt them to an agent system.\n\nYou do not need to understand the whole registry before using it.\n\nWant to experiment first? Open [Try These Five Things](TRY_IT.md).\n\n## Start in 30 seconds\n\n### Use it in a chat\n\nPaste this into a model that can open web links:\n\n```text\nRead https://raw.githubusercontent.com/robkazi52/upgradeables/main/START_HERE.md and follow the model route. Then complete this task: [PASTE YOUR TASK].\n```\n\nIf the model cannot open links, copy [Quick Task](prompts/QUICK_TASK.md), or\nupload [Offline Start](dist/OFFLINE_START.md) plus the one matching compact\nfile from [`dist/recipe-packs/`](dist/recipe-packs/). The\n[all-in-one kit](dist/ALL_IN_ONE_UPGRADEABLE_SKILL_KIT.md) remains available for\ndeep offline inspection, but is intentionally much larger.\n\n### Build a reusable Skill\n\nCopy [Build a Skill](prompts/BUILD_A_SKILL.md) into a new chat. It asks the\nmodel to select components, explain keep/drop decisions, produce a complete\n`SKILL.md`, and define behavioral tests.\n\n### Use it from an IDE or agent\n\nClone the repository. Agents should discover [AGENTS.md](AGENTS.md), while\nGitHub Copilot receives [repository instructions](.github/copilot-instructions.md).\nThe universal model entrypoint is [START_HERE.md](START_HERE.md).\n\n```bash\ngit clone https://github.com/robkazi52/upgradeables.git\ncd upgradeables\npython scripts/query_registry.py --task \"review this pull request\" --brief\npython scripts/query_registry.py --recipe code-review --runtime\n```\n\n### Install it into a project\n\nThe `0.3.0` project harness is local-first and has no runtime dependencies:\n\n```bash\npipx install git+https://github.com/robkazi52/upgradeables.git@v0.3.0\ncd your-project\nupgradeables init\nupgradeables integrate codex --write\nupgradeables task \"review this pull request for regressions\"\n```\n\nThe harness writes its own state under `.upgradeables/`; host instructions are\nchanged only by an explicit integration write. Build a reusable project workflow\nwith `upgradeables skill brief`, `skill scaffold`, and `skill validate`. See the\n[Harness guide](docs/HARNESS.md) and [Project Skill Factory](docs/SKILL_FACTORY.md).\n\n## What is an Upgradeable?\n\nAn Upgradeable is an optional capability with a defined trigger, inputs,\noutputs, procedure, compatibility rules, failure boundary, and tests. Examples\ninclude task locking, source grounding, citation checking, drift control,\nstructured state, and bounded repair.\n\nSkills perform jobs. Upgradeables are building blocks used inside those Skills.\n\n| Layer | What it does |\n|---|---|\n| Skill | Performs a task for a user |\n| Upgradeable | Adds one reusable capability or control |\n| Behavior Gene | Sets a recurring behavior or reasoning style |\n| Core | Supplies domain reasoning and reference knowledge |\n| Validator | Checks, scores, vetoes, or requests repair |\n| Recipe | Suggests a minimal starting composition for a task family |\n| OS / bundle | Coordinates a larger operating environment |\n\n## How to choose Upgradeables\n\nUse one of three short discovery paths:\n\n- **Task → Recipe → Upgradeables:** start in [`recipes/`](recipes/) for a\n  reviewed task-family composition, then retain only components with active triggers.\n- **Upgradeable → Purpose / OS Fit / Tasks / Interactions:** search the compact\n  catalog, inspect a package's placement and exclusions, then follow explained links.\n- **Bundle → Curated multi-Upgradeable composition:** start in [`bundles/`](bundles/)\n  when the task genuinely needs a coordinated subsystem rather than one capability.\n\n## Find the right building blocks\n\n- Human/model router: [START_HERE.md](START_HERE.md)\n- Tiny machine router: [`runtime/router.json`](runtime/router.json)\n- Compact runtime cards and recipe packs: [`runtime/`](runtime/)\n- Compact machine catalog: [`registry/catalog.json`](registry/catalog.json)\n- Full machine registry: [`registry/registry.json`](registry/registry.json)\n- Task-family starting points: [`recipes/`](recipes/)\n- Individual packages: [`upgradeables/`](upgradeables/)\n- Plain-language name recommendations: [v0.2 naming review](audit/NAMING_REVIEW_v0.2.md)\n- Copy-ready chat prompts: [`prompts/`](prompts/)\n- Low-context offline start: [Offline Start](dist/OFFLINE_START.md)\n- Comprehensive offline reference: [all-in-one kit](dist/ALL_IN_ONE_UPGRADEABLE_SKILL_KIT.md)\n- Deeper model instructions: [Model Consumption Guide](MODEL_CONSUMPTION_GUIDE.md)\n\nThe query helper has no third-party dependencies:\n\n```bash\npython scripts/query_registry.py --slug grounding-no-invention\npython scripts/query_registry.py --class validation\npython scripts/query_registry.py --search long-context\npython scripts/query_registry.py --recipe coding-debugging\npython scripts/query_registry.py --task \"triage and fix this GitHub issue\" --brief\npython scripts/query_registry.py --search \"long context\" --brief --limit 5\n```\n\nOr query the compact catalog directly:\n\n```python\nimport json\nfrom pathlib import Path\n\nregistry = json.loads(\n    Path(\"registry/registry.json\").read_text(encoding=\"utf-8\")\n)\n\nresearch = next(\n    recipe[\"classifications\"]\n    for recipe in registry[\"recipes\"]\n    if recipe[\"slug\"] == \"research-skill\"\n)\n```\n\n## Evidence\n\nThe repository now preserves preliminary, author-reported experiments that\nmotivated several composition choices. The ARC report describes one-shot grid\nreasoning runs and a small within-session comparison in which focused directives\nmatched or exceeded a more elaborate prompt on three of five ARC-AGI-2 tasks.\n\nThese reports are not independently reproduced: raw runs and grader artifacts\nare not yet archived, and the supplied ARC-AGI-1 totals contain discrepancies.\nRead the [evidence status and methodology limits](evidence/) before citing any\nnumber. See [Design Principles](DESIGN_PRINCIPLES.md) for the deliberately narrow\nhypotheses the results motivate.\n\n## Build and contribute Skills\n\nUse the [Skill implementation template](templates/SKILL_IMPLEMENTATION_TEMPLATE.md)\nand the [worked community examples](implementations/community/), covering research,\ncoding/debugging, GitHub issue fixing, long-context analysis, creative ideation,\nhigh-stakes evidence, Skill architecture, and ARC-style perception reasoning.\nCommunity members and companies can contribute complete Skills without changing\nthe canonical Upgradeable registry. See [CONTRIBUTING.md](CONTRIBUTING.md) for the\nseparate Skill and Upgradeable contribution paths.\n\n```bash\npython scripts/validate_skill.py implementations/community/source-bounded-research\n```\n\nBefore proposing a new primitive, search the registry and explain why composing\nexisting building blocks is insufficient. Historical names and unresolved ideas\nare preserved rather than guessed or silently rewritten.\n\n## Validate locally\n\nRequires Python 3.11+ and no runtime dependencies:\n\n```bash\npython scripts/build_semantic_packages.py --check\npython scripts/build_ecosystem_reviews.py --check\npython scripts/build_registry.py --check\npython scripts/build_runtime.py --check\npython scripts/build_selection_ontology.py --check\npython scripts/build_harness_data.py --check\npython scripts/validate_registry.py\npython scripts/validate_selection_ontology.py\npython scripts/validate_behavior_cases.py\npython scripts/run_deterministic_package_checks.py\npython scripts/audit_semantic_specificity.py\npython scripts/validate_skill.py\npython -m unittest discover -s tests -v\npython scripts/build_all_in_one.py --check\npython scripts/build_release_assets.py --check\npython scripts/check_links.py\n```\n\n## Give this repo to an LLM\n\nPoint the model to [Start Here](START_HERE.md). It will use the tiny router, an\nexisting Skill, or one compact runtime recipe. For offline use, attach\n[Offline Start](dist/OFFLINE_START.md) and one recipe pack. Use the full registry\nand all-in-one kit only for deep inspection or contribution work.\n\n## Authority and safety\n\nHost and system policy always win. Upgradeables cannot provide hidden memory,\nprivate chain-of-thought access, unavailable tools, or safety bypasses. Models\nmust disclose unavailable capabilities and must not invent unresolved history.\n\nLicensed under [Apache-2.0](LICENSE).\n"}