@bendyline/gilde 0.1.38 → 0.1.40

package/data/chat-models/qw/qwen3.8-27b-q2/manifest.json

@@ -0,0 +1,194 @@
+{
+  "schemaVersion": 1,
+  "kind": "chat-model",
+  "id": "qwen3.8-27b-q2",
+  "name": "Qwen 3.8 (27B, Q2)",
+  "description": "Alibaba's Qwen 3.8 at 27B parameters: a dense, multimodal agentic model with native text, image, and video understanding, 262K context, flexible reasoning effort, MTP, and native tool calling. Unsloth's dynamic UD-Q2_K_XL brings the 27B within reach of machines with 16GB+ usable memory while retaining more fidelity than the extreme IQ1_S build.",
+  "tags": [
+    "alibaba",
+    "agentic",
+    "coding",
+    "multimodal",
+    "vision",
+    "video",
+    "tools",
+    "reasoning",
+    "long-context"
+  ],
+  "category": "general",
+  "maintainer": {
+    "name": "Alibaba",
+    "url": "https://huggingface.co/Qwen/Qwen3.8-27B"
+  },
+  "version": "1.0.0",
+  "updatedAt": "2026-08-23T00:00:00Z",
+  "license": "Apache-2.0",
+  "licenseClass": "open",
+  "licenseShortName": "Apache 2.0",
+  "recoScore": 20,
+  "parameterSize": "27B",
+  "approxSizeBytes": 9828981664,
+  "supportsTools": true,
+  "contextWindow": 262144,
+  "upstream": "https://huggingface.co/Qwen/Qwen3.8-27B",
+  "style": {
+    "family": "qwen",
+    "reasoningFormat": "think",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "reasoning.strip-think-tags",
+    "provider.merge-system-messages",
+    "mcp.compact-tool-schemas",
+    "provider.compact-write-transcript",
+    "fabrication.detect-past-tense-no-tools",
+    "turn.ollama-num-predict-bumped",
+    "turn.preamble-folding",
+    "turn.ramble-detection",
+    "tools.mlx-grammar",
+    "prompt.retrieval-first",
+    "prompt.workspace-gestalt",
+    "prompt.derive-by-execution"
+  ],
+  "tuning": {
+    "sampling": {
+      "temperature": 1,
+      "topP": 0.95,
+      "topK": 20,
+      "minP": 0,
+      "presencePenalty": 0,
+      "repetitionPenalty": 1,
+      "maxTokens": 12288
+    },
+    "samplingWhenThinking": {
+      "temperature": 1,
+      "topP": 0.95
+    },
+    "reasoning": {
+      "enableThinking": true,
+      "thinkingBudget": 4096,
+      "templateKwargs": {
+        "reasoning_effort": "medium"
+      }
+    },
+    "promptTags": {
+      "enableThinkingTag": "/think",
+      "disableThinkingTag": "/no_think"
+    },
+    "profiles": {
+      "thinking-general": {
+        "sampling": {
+          "temperature": 1,
+          "topP": 0.95,
+          "topK": 20,
+          "minP": 0,
+          "presencePenalty": 0,
+          "repetitionPenalty": 1,
+          "maxTokens": 12288
+        },
+        "reasoning": {
+          "enableThinking": true,
+          "thinkingBudget": 4096,
+          "templateKwargs": {
+            "reasoning_effort": "medium"
+          }
+        }
+      },
+      "thinking-deep": {
+        "reasoning": {
+          "enableThinking": true,
+          "thinkingBudget": 4096,
+          "templateKwargs": {
+            "reasoning_effort": "xhigh"
+          }
+        }
+      },
+      "thinking-coding": {
+        "sampling": {
+          "temperature": 1,
+          "topP": 0.95,
+          "topK": 20,
+          "minP": 0,
+          "presencePenalty": 0,
+          "repetitionPenalty": 1,
+          "maxTokens": 16384
+        },
+        "reasoning": {
+          "enableThinking": true,
+          "thinkingBudget": 4096,
+          "templateKwargs": {
+            "reasoning_effort": "medium"
+          }
+        }
+      },
+      "thinking-precise": {
+        "sampling": {
+          "temperature": 0.6,
+          "topP": 0.95,
+          "topK": 20,
+          "minP": 0,
+          "presencePenalty": 0,
+          "repetitionPenalty": 1,
+          "seed": 0,
+          "maxTokens": 6144
+        },
+        "reasoning": {
+          "enableThinking": true,
+          "thinkingBudget": 4096,
+          "templateKwargs": {
+            "reasoning_effort": "medium"
+          }
+        }
+      },
+      "instruct": {
+        "sampling": {
+          "temperature": 0.7,
+          "topP": 0.8,
+          "topK": 20,
+          "minP": 0,
+          "presencePenalty": 1.5,
+          "repetitionPenalty": 1,
+          "maxTokens": 4096
+        },
+        "reasoning": {
+          "enableThinking": false
+        }
+      },
+      "creative": {
+        "sampling": {
+          "temperature": 1,
+          "topP": 0.95,
+          "topK": 40,
+          "minP": 0,
+          "presencePenalty": 1.5,
+          "repetitionPenalty": 1,
+          "maxTokens": 8192
+        },
+        "reasoning": {
+          "enableThinking": false
+        }
+      }
+    },
+    "engine": {
+      "llamaCpp": {
+        "spec": {
+          "mtp": true,
+          "nMax": 4
+        }
+      }
+    }
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "unsloth/Qwen3.8-27B-GGUF",
+    "revision": "4ca720788d1e01f1bff70c033e0d0028fd02e502",
+    "filename": "Qwen3.8-27B-UD-Q2_K_XL.gguf",
+    "sha256": "fd4730dd8aad070517978752b63d530aeb1740d2283cab9fa24f1e404032ddb0",
+    "approxSizeBytes": 9828981664,
+    "quantization": "UD-Q2_K_XL",
+    "mmproj": {
+      "filename": "mmproj-F16.gguf",
+      "sha256": "cbb841a9ee0636b2ec172f5bb8df2ea8dfeb01e90fe7c6126581d662a0b4e43e",
+      "sizeBytes": 927607488
+    }
+  }
+}

package/data/chat-models/qw/qwen3.8-27b-q2/versions/1.0.0/manifest.json

@@ -0,0 +1,19 @@
+{
+  "schemaVersion": 1,
+  "version": "1.0.0",
+  "releasedAt": "2026-08-23T00:00:00Z",
+  "approxSizeBytes": 9828981664,
+  "llamaCpp": {
+    "huggingfaceRepo": "unsloth/Qwen3.8-27B-GGUF",
+    "revision": "4ca720788d1e01f1bff70c033e0d0028fd02e502",
+    "filename": "Qwen3.8-27B-UD-Q2_K_XL.gguf",
+    "sha256": "fd4730dd8aad070517978752b63d530aeb1740d2283cab9fa24f1e404032ddb0",
+    "approxSizeBytes": 9828981664,
+    "quantization": "UD-Q2_K_XL",
+    "mmproj": {
+      "filename": "mmproj-F16.gguf",
+      "sha256": "cbb841a9ee0636b2ec172f5bb8df2ea8dfeb01e90fe7c6126581d662a0b4e43e",
+      "sizeBytes": 927607488
+    }
+  }
+}

package/data/craftbook-templates/a1/a11y-audit/manifest.json

@@ -3,6 +3,7 @@
   "kind": "craftbook-template",
   "id": "a11y-audit",
   "role": "maintenance-review",
+  "category": "code-review",
   "name": "Accessibility Audit",
   "description": "Audit a web page or UI for accessibility against WCAG 2.1 AA and produce a prioritized remediation report. First scopes the components and flows in play and locks a WCAG checklist (semantic structure, keyboard operability, focus management, color contrast, alt text, form labels, ARIA correctness, motion/reflow), then audits each criterion by inspecting the markup, simulating keyboard and screen-reader use, and measuring contrast, then writes a findings report citing the failing WCAG success criterion, the affected element, and the fix. Use this for an accessibility audit, a11y review, WCAG compliance check, screen-reader or keyboard-navigation testing, or making a page accessible — it ties each defect to a numbered WCAG criterion with a concrete code remedy rather than vague 'improve accessibility' notes.",
   "tags": [

package/data/craftbook-templates/a1/a11y-audit/versions/1.1.2/craftbook.json

@@ -0,0 +1,153 @@
+{
+  "id": "a11y-audit",
+  "name": "Accessibility Audit",
+  "description": "Audit a web page or UI for accessibility against WCAG 2.1 AA and produce a prioritized remediation report. First scopes the components and flows in play and locks a WCAG checklist (semantic structure, keyboard operability, focus management, color contrast, alt text, form labels, ARIA correctness, motion/reflow), then audits each criterion by inspecting the markup, simulating keyboard and screen-reader use, and measuring contrast, then writes a findings report citing the failing WCAG success criterion, the affected element, and the fix. Use this for an accessibility audit, a11y review, WCAG compliance check, screen-reader or keyboard-navigation testing, or making a page accessible — it ties each defect to a numbered WCAG criterion with a concrete code remedy rather than vague 'improve accessibility' notes.\n\nA gallery craftbook generated from an archetype spec. It runs\n`phase → (per-phase gate) → … → evaluate → (loop) → finish`. Each build\nphase that produces a checkable artifact is followed by a **runtime\ngate-checkpoint** — the runtime verifies the artifact and routes with no\nmodel turn, looping back to redo the phase on a miss. The final `evaluate`\nstep holds a static deliverable gate plus a reviewer QA pass. What it adds\nover the generic `build-loop`: a specialist role per phase, a\ndomain-correct ordering, and a concrete per-phase quality bar.\n\nDeliverables marked \"artifact\" land in the project's artifacts drawer (`write_artifact` / `read_artifact`), not the shipped workspace — review output is not product source.\n\nPhases:\n\n1. Scope the audit (reviewer) — enumerate components + lock a WCAG AA checklist → gated on artifact `{{workPath}}/scope.md` (markdown-notes)\n2. Audit each criterion (reviewer) — inspect markup, keyboard, contrast, ARIA per SC → gated on artifact `{{workPath}}/audit.md` (markdown-notes)\n3. Write the a11y report (reviewer) — findings keyed to WCAG SCs with fixes → gated on artifact `{{workPath}}/a11y-report.md` (markdown-report)\n\nThe gates never advance with an unmet criterion, and loop back to the\nowning phase to fix named gaps.\n",
+  "entryStepId": "scope",
+  "triggers": [
+    "accessibility audit",
+    "a11y review",
+    "check WCAG compliance",
+    "is this accessible",
+    "screen reader test",
+    "keyboard navigation audit"
+  ],
+  "steps": [
+    {
+      "id": "scope",
+      "name": "Scope the audit",
+      "description": "enumerate components + lock a WCAG AA checklist",
+      "prompt": "Step 1: Identify the page/components and the user flows to audit (navigation, forms, modals, interactive widgets, media). Step 2: Build a WCAG 2.1 AA acceptance-criteria checklist covering: semantic landmarks and heading order (1.3.1), keyboard operability of all controls with no traps (2.1.1/2.1.2), visible focus indicators and logical focus order (2.4.7/2.4.3), text contrast ≥ 4.5:1 and UI/large text ≥ 3:1 (1.4.3/1.4.11), text alternatives for images/icons (1.1.1), programmatic labels for every form field (1.3.1/4.1.2), correct and minimal ARIA (4.1.2), name/role/value for custom widgets, reflow at 320px and 200% zoom (1.4.10/1.4.4), and no information conveyed by color alone (1.4.1). Step 3: For each item, state the pass condition for THIS UI. Step 4: Note the tools/method you will use (DOM inspection, keyboard walk, contrast math, axe if available). Write the component list + WCAG checklist (each item tagged with its SC number) to `write_task_note` AND `{{workPath}}/scope.md`. No findings yet.\n\nThe deliverable `{{workPath}}/scope.md` lands in the project's artifacts drawer — write it with `write_artifact` and read it back with `read_artifact`; the shipped workspace stays untouched.",
+      "suggestedRole": "reviewer",
+      "advanceWhen": {
+        "file": "{{workPath}}/scope.md",
+        "minBytes": 1,
+        "sniff": "nonempty",
+        "artifact": true
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "{{workPath}}/scope.md",
+            "bytes": 120,
+            "artifact": true
+          },
+          {
+            "kind": "sniff",
+            "file": "{{workPath}}/scope.md",
+            "sniff": "nonempty",
+            "artifact": true
+          }
+        ],
+        "onReject": "scope",
+        "maxAttempts": 3
+      },
+      "next": "audit"
+    },
+    {
+      "id": "audit",
+      "name": "Audit each criterion",
+      "description": "inspect markup, keyboard, contrast, ARIA per SC",
+      "prompt": "Step 1: Open/inspect the UI (use the browser or a11y tooling if available; otherwise read the HTML/CSS/JS and reason about the rendered semantics). Step 2: For EACH checklist criterion, perform the actual check: tab through the page recording focus order and any unreachable/trapped controls; verify each interactive element has an accessible name and correct role; compute contrast ratios for text and UI on their real backgrounds; confirm every image/icon has appropriate alt or is correctly hidden; confirm every input has a programmatic label; validate ARIA usage against the spec (no redundant/invalid roles). Step 3: For each violation capture: the WCAG SC number and level, the selector/element, what is wrong, the user impact (who is blocked and how), severity, and the concrete code fix. Step 4: Record the criteria that PASS so the report shows full coverage. Good looks like: 'SC 1.4.3 FAIL — .btn-primary text #9aa on #fff = 2.1:1, needs ≥4.5:1; darken to #555'. Stage findings to `write_task_note`. Do not write the report yet.\n\nThe deliverable `{{workPath}}/audit.md` lands in the project's artifacts drawer — write it with `write_artifact` and read it back with `read_artifact`; the shipped workspace stays untouched.",
+      "suggestedRole": "reviewer",
+      "advanceWhen": {
+        "file": "{{workPath}}/audit.md",
+        "minBytes": 1,
+        "sniff": "nonempty",
+        "artifact": true
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "{{workPath}}/audit.md",
+            "bytes": 120,
+            "artifact": true
+          },
+          {
+            "kind": "sniff",
+            "file": "{{workPath}}/audit.md",
+            "sniff": "nonempty",
+            "artifact": true
+          }
+        ],
+        "onReject": "audit",
+        "maxAttempts": 3
+      },
+      "next": "report"
+    },
+    {
+      "id": "report",
+      "name": "Write the a11y report",
+      "description": "findings keyed to WCAG SCs with fixes",
+      "prompt": "Step 1: Open `{{workPath}}/a11y-report.md` and write a top-line verdict (Conformant / Partially-conformant / Non-conformant to WCAG 2.1 AA) with counts by severity. Step 2: Add a '## Findings' section; for each, write `### [SEVERITY] SC <number> (<Level>) — <element>` with what is wrong, the user impact, and the exact code fix. Order critical-first. Step 3: Add a '## Passing criteria' section listing the SCs checked and met, so coverage is auditable. Step 4: Add a '## Manual verification needed' section for items a static check cannot fully settle (e.g. screen-reader announcement quality). Every finding must cite a specific SC and element with a concrete remedy — no generic 'add ARIA'. On a loop-back, fix only the named gaps. `write_task_note` the report path and the severity counts.\n\nThe deliverable `{{workPath}}/a11y-report.md` lands in the project's artifacts drawer — write it with `write_artifact` and read it back with `read_artifact`; the shipped workspace stays untouched.",
+      "suggestedRole": "reviewer",
+      "advanceWhen": {
+        "file": "{{workPath}}/a11y-report.md",
+        "minBytes": 1,
+        "sniff": "nonempty",
+        "artifact": true
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "{{workPath}}/a11y-report.md",
+            "bytes": 1500,
+            "artifact": true
+          },
+          {
+            "kind": "contains",
+            "file": "{{workPath}}/a11y-report.md",
+            "pattern": "(?:^|\\n)#{1,3}\\s+\\S",
+            "flags": "i",
+            "label": "at least one markdown heading",
+            "artifact": true
+          }
+        ],
+        "onReject": "report",
+        "maxAttempts": 4
+      },
+      "next": "evaluate"
+    },
+    {
+      "id": "evaluate",
+      "name": "Evaluate",
+      "description": "Grade the deliverable against every acceptance criterion. All pass → finish; any fail → loop back and fix the gap.",
+      "prompt": "Open `{{workPath}}/a11y-report.md` and grade it against the scope WCAG checklist. Check EACH criterion and write PASS/FAIL with a one-line reason: (1) every checklist SC was assessed (found-and-reported or marked passing); (2) each finding cites a specific WCAG success-criterion number and level; (3) each finding names the affected element and a concrete code fix; (4) contrast findings include actual computed ratios; (5) keyboard operability and focus order were actually walked; (6) a passing-criteria section makes coverage auditable; (7) the verdict matches the findings. If any criterion fails, name the exact gap and loop back.\n\nThe deliverable lives in the project's artifacts drawer — open `{{workPath}}/a11y-report.md` with `read_artifact`, not `read_file`.\n\nThen route — this is the whole point of the loop:\n\n- **Every criterion PASSES →** call `advance_task_step({ ref, stepId: \"evaluate\", next: \"finish\" })`.\n- **Any criterion FAILS →** write the specific gaps to notes, then call `advance_task_step({ ref, stepId: \"evaluate\", next: \"report\" })` to loop back. The builder fixes exactly those gaps.\n\nNever route to `finish` while any criterion is unmet. The build phase's completion gate already blocked a grossly-incomplete deliverable; your job is the judgment an automated check cannot make (does it actually work, read well, look right). After ~3 unproductive loops, stop and report DONE_WITH_CONCERNS so the user can step in.",
+      "suggestedRole": "reviewer",
+      "consumes": [
+        {
+          "file": "{{workPath}}/a11y-report.md",
+          "artifact": true
+        }
+      ],
+      "next": "report"
+    },
+    {
+      "id": "finish",
+      "name": "Finish",
+      "description": "All acceptance criteria met. Stamp a short summary and report DONE.",
+      "prompt": "Every acceptance criterion passed. Write a one-paragraph DONE summary to task notes via `write_task_note`: what was built, the deliverable path(s), and a one-line confirmation that each criterion is met. Then report DONE.",
+      "suggestedRole": "developer",
+      "terminal": true
+    }
+  ],
+  "version": "1.1.2",
+  "releasedAt": "2026-08-21T00:00:00Z",
+  "paramSchema": {
+    "type": "object",
+    "properties": {
+      "workPath": {
+        "type": "string",
+        "title": "Working folder",
+        "description": "Per-task working folder in the artifacts drawer. Defaults to this task's own folder so runs never collide; override with a stable name when you deliberately want runs to share files.",
+        "default": "{{task.dir}}"
+      }
+    }
+  },
+  "minGezelVersion": "1.26233"
+}

package/data/craftbook-templates/a1/a11y-audit/versions/1.1.2/test.json

@@ -0,0 +1,118 @@
+{
+  "schemaVersion": 1,
+  "title": "Accessibility Audit codebase eval",
+  "objective": "Measure whether the audit finds concrete WCAG failures across a seeded UI and leaves a source-cited remediation report.",
+  "tags": [
+    "corpus"
+  ],
+  "prompt": "Use the Accessibility Audit craftbook for a whole-codebase maintenance pass. Review the seeded UI against WCAG 2.1 AA and write tasks/eval/a11y-report.md. Do not change the UI.",
+  "setup": {
+    "projectName": "Accessibility Maintenance Eval",
+    "about": "A tiny dashboard UI with deliberate accessibility defects. Source files are authoritative and no network access is needed.",
+    "missionObjectives": "Find the seeded keyboard, labeling, semantic, image-text, and contrast defects; cite real files/elements and propose exact fixes in tasks/eval/a11y-report.md.",
+    "files": [
+      {
+        "path": "src/App.tsx",
+        "content": "export function App() {\n  return <main>\n    <h1>Returns desk</h1>\n    <h3>Find an order</h3>\n    <img src=\"/returns.png\" />\n    <div className=\"lookup\" onClick={() => location.assign('/lookup')}>Look up order</div>\n    <input placeholder=\"Order number\" />\n    <div role=\"button\">Export exceptions</div>\n  </main>;\n}\n"
+      },
+      {
+        "path": "src/styles.css",
+        "content": "body { background: #fff; color: #aaa; }\n.lookup { color: #aaa; background: #fff; }\n*:focus { outline: none; }\n"
+      }
+    ],
+    "worker": {
+      "name": "Rae",
+      "role": "Reviewer"
+    },
+    "craftbookParams": {
+      "workPath": "tasks/eval"
+    }
+  },
+  "mocks": [],
+  "success": {
+    "summary": "tasks/eval/a11y-report.md cites the seeded UI and covers every deliberate WCAG failure with a concrete fix.",
+    "deliverables": [
+      {
+        "path": "tasks/eval/a11y-report.md",
+        "kind": "markdown-report",
+        "minBytes": 1300,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "tasks/eval/a11y-report.md",
+            "pattern": "src/App\\.tsx|src/styles\\.css",
+            "flags": "i",
+            "label": "cite real source"
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/a11y-report.md",
+            "pattern": "1\\.1\\.1",
+            "flags": "i",
+            "label": "missing image alternative"
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/a11y-report.md",
+            "pattern": "1\\.3\\.1|4\\.1\\.2",
+            "flags": "i",
+            "label": "labels and semantics"
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/a11y-report.md",
+            "pattern": "2\\.1\\.1|keyboard",
+            "flags": "i",
+            "label": "keyboard operability"
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/a11y-report.md",
+            "pattern": "1\\.4\\.3|contrast",
+            "flags": "i",
+            "label": "contrast"
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/a11y-report.md",
+            "pattern": "#aaa|ratio|4\\.5:1",
+            "flags": "i",
+            "label": "measured contrast evidence"
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/a11y-report.md",
+            "pattern": "outline|focus",
+            "flags": "i",
+            "label": "focus visibility"
+          }
+        ]
+      }
+    ]
+  },
+  "rubric": {
+    "artifact": {
+      "path": "tasks/eval/a11y-report.md",
+      "kind": "markdown"
+    },
+    "axes": [
+      {
+        "name": "coverage",
+        "description": "The seeded WCAG failures and passing/unknown criteria are accounted for."
+      },
+      {
+        "name": "evidence",
+        "description": "Findings cite concrete elements, source paths, and measured values."
+      },
+      {
+        "name": "remediation",
+        "description": "Each issue includes a specific code-level fix."
+      }
+    ]
+  },
+  "qualityFocus": [
+    "WCAG-specific findings",
+    "source evidence",
+    "concrete remediation"
+  ]
+}

package/data/craftbook-templates/ab/ab-test-readout/manifest.json

@@ -3,6 +3,7 @@
   "kind": "craftbook-template",
   "id": "ab-test-readout",
   "role": "general",
+  "category": "research",
   "name": "A/B Test Readout",
   "description": "Analyze a controlled experiment and produce a decision-ready readout: does the variant beat control, by how much, and is it real? States the hypothesis, the primary metric, and the decision rule FIRST (to avoid p-hacking and metric-shopping), then computes lift, statistical significance, and confidence intervals as a data-analyst, then writes a plain-language ship/no-ship recommendation. Use this for an experiment analysis, A/B test results, significance testing, conversion-lift readout, or an experiment decision memo.",
   "tags": [