@sanity/ailf 0.4.1 → 1.0.0

package/config/features.yaml

@@ -1,116 +0,0 @@
-# features.yaml
-#
-# Product feature registry for documentation coverage auditing.
-# Each entry represents a product feature that should have evaluation
-# task coverage. Edit this file to track coverage gaps and priorities.
-#
-# The coverage audit (`pnpm coverage-audit`) cross-references this
-# registry against tasks/*.yaml to identify covered and uncovered features.
-#
-# Status values:
-#   covered      — has evaluation tasks in tasks/*.yaml
-#   uncovered    — no evaluation tasks yet
-#   planned      — tasks are planned but not yet written
-#   out-of-scope — intentionally excluded from evaluation
-#
-# Phase 3c of the Scenario Matrix implementation.
-# See docs/exec-plans/scenario-matrix-implementation/phase-3-gap-analysis.md
-
-features:
-  # === Currently covered (have evaluation tasks) ===
-  - id: groq
-    name: "GROQ Query Language"
-    sections: [content-lake]
-    status: covered
-    area: groq
-    priority: critical
-    taskCount: 3
-
-  - id: visual-editing
-    name: "Visual Editing"
-    sections: [visual-editing]
-    status: covered
-    area: visual-editing
-    priority: critical
-    taskCount: 1
-
-  - id: nextjs-live
-    name: "Next.js Live Preview"
-    sections: [visual-editing]
-    status: covered
-    area: nextjs-live
-    priority: high
-    taskCount: 2
-
-  - id: functions
-    name: "Sanity Functions"
-    sections: [compute-and-ai]
-    status: covered
-    area: functions
-    priority: high
-    taskCount: 2
-
-  - id: studio-setup
-    name: "Studio Configuration"
-    sections: [studio]
-    status: covered
-    area: studio-setup
-    priority: high
-    taskCount: 1
-
-  - id: frameworks
-    name: "Framework Integration"
-    sections: [developer-guides]
-    status: covered
-    area: frameworks
-    priority: high
-    taskCount: 2
-
-  # === Uncovered (no evaluation tasks yet) ===
-  - id: portable-text
-    name: "Portable Text"
-    sections: [content-lake, studio]
-    status: uncovered
-    priority: high
-
-  - id: image-assets
-    name: "Image & Asset Handling"
-    sections: [content-lake, apis-and-sdks]
-    status: uncovered
-    priority: high
-
-  - id: mutations
-    name: "Mutations & Transactions"
-    sections: [content-lake, apis-and-sdks]
-    status: uncovered
-    priority: high
-
-  - id: schemas
-    name: "Schema Types & Validation"
-    sections: [studio, content-lake]
-    status: uncovered
-    priority: medium
-
-  - id: authentication
-    name: "Authentication & Access Control"
-    sections: [apis-and-sdks]
-    status: uncovered
-    priority: medium
-
-  - id: webhooks
-    name: "Webhooks"
-    sections: [content-lake]
-    status: uncovered
-    priority: medium
-
-  - id: realtime
-    name: "Real-time Listeners"
-    sections: [apis-and-sdks]
-    status: uncovered
-    priority: low
-
-  - id: ai-assist
-    name: "AI Assist"
-    sections: [compute-and-ai, studio]
-    status: uncovered
-    priority: medium

package/config/models.yaml

@@ -1,116 +0,0 @@
-# models.yaml
-#
-# Central model registry for ai-literacy-framework evaluations.
-#
-# Define all models you want to test here. Each eval mode (baseline, observed,
-# agentic) reads this file and generates the appropriate provider entries.
-#
-# Usage:
-#   1. Add/remove models below
-#   2. Run: pnpm generate-configs
-#   3. Run your eval: pnpm eval / pnpm eval:observed / pnpm eval:agentic
-#
-# Model entries support:
-#   - id:          Promptfoo provider identifier (e.g., "openai:gpt-4o")
-#   - label:       Human-readable label for results display
-#   - config:      Model-specific config (temperature, max_tokens, etc.)
-#   - modes:       Which eval modes to include this model in (default: all)
-#                  Options: baseline, observed, agentic-naive, agentic-optimized
-#   - env:         Environment variable name for the API key (auto-detected for
-#                  openai:* and anthropic:* providers)
-#
-# The generator script expands each model into the correct provider format
-# for each eval mode. Custom providers (observed, agentic) are wired up
-# automatically — you just specify the model name.
-
-models:
-  # -- Anthropic
-  - id: anthropic:messages:claude-opus-4-6
-    label: Claude Opus 4.6
-    config:
-      temperature: 0.2
-      max_tokens: 4096
-    modes: [baseline, observed, agentic-naive, agentic-optimized]
-  # -- Google
-  # - id: google:gemini-2.5-pro
-  #   label: Gemini 2.5 Pro
-  #   config:
-  #     temperature: 0.2
-  #     max_tokens: 4096
-  #   modes: [baseline, observed, agentic-naive, agentic-optimized]
-  # -- OpenAI
-  - id: openai:chat:gpt-5.2
-    label: GPT 5.2
-    config:
-      temperature: 0.2
-      max_tokens: 4096
-    modes: [baseline, observed, agentic-naive, agentic-optimized]
-  - id: openai:chat:gpt-5.4
-    label: GPT 5.4
-    config:
-      reasoning_effort: "medium"
-      max_output_tokens: 4096
-      maxRetries: 1
-    modes: [baseline, observed, agentic-naive, agentic-optimized]
-
-  # ── Anthropic ───────────────────────────────────────────────
-  # - id: anthropic:claude-sonnet-4-20250514
-  #   label: Claude Sonnet 4
-  #   config:
-  #     temperature: 0.2
-  #     max_tokens: 4096
-  #   modes: [baseline]
-
-  # - id: anthropic:claude-3.5-sonnet-20241022
-  #   label: Claude 3.5 Sonnet
-  #   config:
-  #     temperature: 0.2
-  #     max_tokens: 4096
-  #   modes: [baseline, agentic-naive, agentic-optimized]
-
-  # ── Google ──────────────────────────────────────────────────
-  # - id: google:gemini-2.0-flash
-  #   label: Gemini 2.0 Flash
-  #   config:
-  #     temperature: 0.2
-  #     max_tokens: 4096
-  #   modes: [baseline]
-
-  # ── Other ───────────────────────────────────────────────────
-  # - id: openrouter:deepseek/deepseek-r1
-  #   label: DeepSeek R1
-  #   config:
-  #     temperature: 0.2
-  #     max_tokens: 4096
-  #   modes: [baseline]
-
-# ── Grading Model ───────────────────────────────────────────
-# Which model scores the responses. Separate from the models being tested.
-grader:
-  id: anthropic:messages:claude-opus-4-5-20251101
-  label: Claude Opus 4.5 (grader)
-#grader:
-#  id: openai:gpt-5-2025-08-07
-#  label: GPT-5 (grader)
-
-# ── Evaluation Options ──────────────────────────────────────
-# Controls how promptfoo runs evaluations.
-maxConcurrency: 32 # max parallel API calls — benchmarked in DOC-1896
-
-# ── Default Config ──────────────────────────────────────────
-# Applied to all models unless overridden per-model.
-defaults:
-  temperature: 0.2
-  max_tokens: 4096
-  maxToolRounds: 5 # for agentic modes
-  observerOptions:
-    maxPreviewBytes: 2048
-    captureResponsePreview: true
-    includePatterns:
-      - "sanity.io"
-      - "sanity.dev"
-      - "cdn.sanity.io"
-    sensitiveHeaders:
-      - "authorization"
-      - "cookie"
-      - "x-api-key"

package/config/prompts.yaml

@@ -1,75 +0,0 @@
-# prompts.yaml
-#
-# Prompt templates used across all evaluation modes.
-# Edit these to change what instructions the LLM receives.
-#
-# Available template variables (injected from task vars):
-#   {{task}}  — the implementation task description
-#   {{docs}}  — documentation context (empty string for baseline tests)
-#
-# Each prompt has:
-#   id:       unique identifier (used in Promptfoo config)
-#   label:    human-readable name (shown in results)
-#   template: the actual prompt text with {{variable}} placeholders
-
-with-docs:
-  id: with-docs
-  label: With Documentation
-  template: |
-    You are an expert Sanity.io developer. Use the following documentation to help implement the task.
-
-    ## Sanity Documentation
-    {{docs}}
-
-    ## Task
-    {{task}}
-
-    ## Requirements
-
-    1. Use ONLY the APIs and patterns shown in the documentation
-    2. Provide a complete, working implementation
-    3. Include all necessary imports
-    4. Follow Sanity best practices as documented
-
-    Provide your implementation:
-
-without-docs:
-  id: without-docs
-  label: Baseline (No Docs)
-  template: |
-    You are an expert Sanity.io developer.
-
-    ## Task
-    {{task}}
-
-    ## Requirements
-
-    1. Provide a complete, working implementation
-    2. Include all necessary imports
-    3. Follow Sanity best practices
-
-    Provide your implementation:
-
-agentic:
-  id: agentic
-  label: Agentic (self-retrieval)
-  template: |
-    You are an expert developer helping implement a Sanity.io feature.
-    You have access to web search and page fetching tools.
-
-    IMPORTANT: Before writing any code, search for and read the relevant
-    Sanity.io documentation to ensure you are using the latest APIs and
-    best practices. Do not rely on memory alone.
-
-    ## Task
-    {{task}}
-
-    ## Requirements
-
-    1. Search for relevant Sanity documentation before implementing
-    2. Use ONLY the APIs and patterns from the current official docs
-    3. Provide a complete, working implementation
-    4. Include all necessary imports
-    5. Follow Sanity best practices as documented
-
-    Provide your implementation:

package/config/rubrics.yaml

@@ -1,62 +0,0 @@
-# rubrics.yaml
-#
-# Centralized rubric templates for LLM grading assertions.
-# Tasks reference these templates by key and provide only their
-# unique criteria bullet points. The pipeline assembles the full
-# rubric text at expansion time.
-#
-# Each dimension is scored on a uniform 0–100 scale. Dimensions are
-# combined into a composite score using the weights below.
-#
-# Each template carries a `dimension` field that tags the scoring
-# dimension it belongs to. This metadata propagates through the
-# expansion pipeline into Promptfoo assertion metadata, enabling
-# structured dimension classification at scoring time instead of
-# fragile heuristic string matching.
-# See docs/design-docs/structured-dimensions.md for the design.
-#
-# See docs/exec-plans/rubric-templates.md for the design.
-# See docs/design-docs/uniform-dimension-scoring.md for the scoring model.
-
-templates:
-  task-completion:
-    dimension: task-completion
-    header: "Score task completion from 0 to 100:"
-    scale:
-      - "0: Couldn't attempt — missing critical information"
-      - "20: Attempted but fundamentally wrong approach"
-      - "50: Partial implementation — major functional gaps"
-      - "80: Mostly complete — minor issues or missing edge cases"
-      - "100: Fully functional code — works as expected"
-    criteria_label: "Must demonstrate:"
-
-  code-correctness:
-    dimension: code-correctness
-    header: "Score code correctness from 0 to 100:"
-    scale:
-      - "0: Broken code, syntax errors, or deprecated APIs"
-      - "30: Works but uses anti-patterns or inefficient approaches"
-      - "50: Works but not idiomatic"
-      - "80: Follows most best practices"
-      - "100: Follows all best practices, idiomatic implementation"
-    criteria_label: "Check for:"
-
-  doc-coverage:
-    dimension: doc-coverage
-    header: "Score documentation coverage from 0 to 100:"
-    scale:
-      - "0: Had to hallucinate/guess most implementation details"
-      - "30: Significant gaps — filled with assumptions"
-      - "50: Some gaps — inferred from partial information"
-      - "80: Minor gaps — almost everything was documented"
-      - "100: Complete coverage — all necessary info was in docs"
-
-# Weights for combining dimension scores into a composite (must sum to 1.0).
-# Composite = task-completion × 0.50 + code-correctness × 0.25 + doc-coverage × 0.25
-weights:
-  task-completion: 0.50
-  code-correctness: 0.25
-  doc-coverage: 0.25
-
-footer:
-  'Return ONLY a JSON object: {"score": <number>, "reason": "<explanation>"}'

package/config/schedules.yaml

@@ -1,43 +0,0 @@
-# schedules.yaml
-#
-# Scheduled evaluation configuration for ai-literacy-framework.
-# Each schedule defines a recurring pipeline run with its own source,
-# mode, and delivery preferences.
-#
-# The GitHub Actions cron workflow (.github/workflows/scheduled-eval.yml)
-# reads this file to determine which evaluations to run and when.
-#
-# Schedule names are used as tags on published reports for easy filtering.
-# Cron expressions use UTC timezone (GitHub Actions standard).
-#
-# @see docs/design-docs/report-store/implementation.md — Phase 5
-
-schedules:
-  # Daily baseline — track score trends against production docs
-  - name: daily-baseline
-    cron: "0 2 * * *" # 2:00 AM UTC, every day
-    mode: baseline
-    source: production
-    publish: true
-    compare: true
-    enabled: true
-
-  # Weekly full decomposition — complete floor/ceiling/actual report
-  # Runs both baseline + agentic in a single pipeline invocation,
-  # producing the three-layer decomposition with retrieval gap metrics.
-  # Replaces the previous weekly-full (baseline) + weekly-agentic schedule pair.
-  - name: weekly-full
-    cron: "0 3 * * 0" # 3:00 AM UTC, every Sunday
-    mode: full
-    source: production
-    publish: true
-    compare: true
-    enabled: true
-
-# Digest configuration — aggregates reports into periodic summaries
-digest:
-  # Weekly digest — summarize score trends from the past 7 days
-  enabled: true
-  cron: "0 9 * * 1" # 9:00 AM UTC, every Monday
-  lookbackDays: 7
-  slackWebhookUrl: ${{ SLACK_WEBHOOK_URL }}

package/config/sinks.yaml

@@ -1,54 +0,0 @@
-# sinks.yaml
-#
-# Report delivery sink configuration for ai-literacy-framework.
-# Sinks receive published evaluation reports and deliver them to external
-# systems (BigQuery, Slack, GitHub, webhooks, etc.).
-#
-# Sinks are fire-and-forget (P6): a sink failure is logged but never blocks
-# the pipeline. The Sanity Content Lake is the system of record.
-#
-# Environment variables use ${{ VAR }} or ${{ VAR | default }} syntax.
-# A sink with `enabled: false` (or a falsy env var) is skipped entirely.
-#
-# Sinks activate only when their required environment variables are present.
-# A developer running locally with no env vars gets zero sinks.
-#
-# @see docs/design-docs/report-store/sink-architecture.md
-
-sinks:
-  # BigQuery — disabled; Airbyte ELT is the primary BigQuery delivery mechanism.
-  # The BigQuerySink can be re-enabled as a fallback if Airbyte is unavailable.
-  # See config/airbyte/ for the active connector and config/bigquery/ for views.
-  # - type: bigquery
-  #   enabled: false
-  #   project: ${{ BIGQUERY_PROJECT | data-platform-302218 }}
-  #   dataset: ${{ BIGQUERY_DATASET | ailf }}
-  #   credentials: ${{ GOOGLE_APPLICATION_CREDENTIALS }}
-
-  # Slack — regression alerts to configured channels
-  # Activates only when SLACK_WEBHOOK_URL is set in the environment.
-  # By default only posts on regressions (avoids notification fatigue).
-  # Routing controls which messages go where by severity level.
-  # - type: slack
-  #   enabled: true
-  #   webhookUrl: ${{ SLACK_WEBHOOK_URL }}
-  #   channel: "#docs-ai-literacy"
-  #   routing:
-  #     critical: "#docs-alerts"    # critical violations → dedicated alert channel
-  #     warning: "#docs-team"       # warnings → team channel
-  #     regression: "#docs-team"    # regressions → team channel
-  #     digest: "#docs-weekly"      # weekly digests → broader channel
-
-  # GitHub PR comments — score tables on CI-triggered eval runs
-  # - type: github-comment
-  #   enabled: false
-  #   token: ${{ GITHUB_TOKEN }}
-
-  # Webhook — generic HTTP relay for Airbyte, Zapier, custom services
-  # - type: webhook
-  #   enabled: false
-  #   url: ${{ AILF_WEBHOOK_URL }}
-  #   headers:
-  #     Authorization: "Bearer ${{ AILF_WEBHOOK_TOKEN }}"
-  #   routing:
-  #     critical: true              # webhooks fire on critical only

package/config/sources.yaml

@@ -1,51 +0,0 @@
-# sources.yaml
-#
-# Documentation source definitions for ai-literacy-framework evaluations.
-# Each source defines where to find documentation for both evaluation modes:
-#
-#   - Baseline/Observed: reads from Sanity CMS (projectId + dataset)
-#   - Agentic: reads from a live URL (baseUrl, llms.txt derived automatically)
-#
-# Environment variables use ${{ VAR }} or ${{ VAR | default }} syntax.
-# Required vars (no default) will error if not set at load time.
-#
-# Available fields (all optional except baseUrl, projectId, dataset):
-#   projectId:      Sanity project ID
-#   dataset:        Sanity dataset name
-#   baseUrl:        Documentation site URL
-#   perspective:    Sanity release perspective ID (for evaluating content releases)
-#   studioOrigin:   Sanity Studio base URL (default: https://admin.sanity.io)
-#   allowedOrigins: Origin allowlist for agentic sandboxing
-#
-# Fields settable only via CLI flags or env vars (not in this file):
-#   documentIds:   --sanity-document / SANITY_DOCUMENT_IDS
-#   urls:          --url / DOC_BASE_URL
-#   headers:       --header / DOC_HEADERS
-#
-# Usage:
-#   pnpm eval --source production
-#   pnpm eval:agentic --source branch
-#   DOC_BASE_URL=https://my-branch.sanity.dev/docs pnpm eval:agentic
-
-sources:
-  # Production Sanity docs — the default when no --source is specified
-  production:
-    projectId: ${{ SANITY_PROJECT_ID | 3do82whm }}
-    dataset: ${{ SANITY_DATASET | next }}
-    baseUrl: https://www.sanity.io/docs
-
-  # Branch deploy — for testing doc changes before merge
-  # Requires DOC_BASE_URL to be set in the environment
-  # Headers can also be passed via --header flag or DOC_HEADERS env var
-  branch:
-    projectId: ${{ SANITY_PROJECT_ID | 3do82whm }}
-    dataset: ${{ SANITY_DATASET | next }}
-    baseUrl: ${{ DOC_BASE_URL }}
-    allowedOrigins:
-      - ${{ DOC_ALLOWED_ORIGIN | }}
-
-  # Local dev server — serve docs locally for offline testing
-  local:
-    projectId: ${{ SANITY_PROJECT_ID | 3do82whm }}
-    dataset: ${{ SANITY_DATASET | next }}
-    baseUrl: http://localhost:${{ DOC_LOCAL_PORT | 3001 }}/docs

package/config/thresholds.yaml

@@ -1,49 +0,0 @@
-# thresholds.yaml
-#
-# Quality thresholds for readiness gates and regression alerts.
-# Each threshold defines a minimum acceptable score. Violations are
-# classified by severity and routed to configured sinks.
-#
-# Used by:
-# - `pnpm pipeline --readiness` (launch readiness checklist)
-# - `pnpm pipeline --publish` (severity-aware sink routing)
-# - `pnpm pipeline --compare` (regression alerting)
-#
-# @see docs/exec-plans/scenario-matrix-implementation/phase-5-readiness-thresholds.md
-
-# Global defaults (apply to all areas unless overridden)
-defaults:
-  composite: 50 # minimum composite score
-  dimensions:
-    task-completion: 40
-    code-correctness: 30
-    doc-coverage: 30
-  doc-lift: 0 # minimum Doc Lift (0 = docs must not hurt)
-  ceiling: 40 # minimum ceiling score (doc quality floor)
-
-# Per-area overrides (inherit from defaults, override specific values)
-areas:
-  groq:
-    composite: 60 # GROQ is critical — higher bar
-    dimensions:
-      task-completion: 50
-  # visual-editing:
-  #   composite: 45 # currently at 36, set achievable near-term target
-  # Areas not listed here use defaults
-
-# Regression thresholds (for comparison reports)
-regression:
-  composite: -3 # alert if composite drops more than 3 points
-  per-area: -5 # alert if any area drops more than 5 points
-  per-dimension: -8 # alert if any dimension drops more than 8 points
-
-# Severity classification
-severity:
-  critical: # blocks deployment, immediate notification
-    composite-below: 30
-    negative-doc-lift: true
-  warning: # flags for review, non-blocking
-    composite-below: 50
-    regression-exceeds: -3
-  info: # logged but not alerted
-    composite-below: 60

package/dist/agent-observer/test-imports.d.ts

@@ -1,7 +0,0 @@
-/**
- * Final validation — ensures all agent-observer modules work together
- * and the full data pipeline (record → classify → summarize) is correct.
- *
- * Run: tsx src/agent-observer/test-imports.ts
- */
-export {};