ivd

IVD Recipes

Reusable patterns for common development scenarios.

agent-rules-ivd.yaml — IVD verification rules for agent instruction files (.cursorrules, .clinerules, Copilot system prompts). Embeds the full 6-rule IVD workflow: intent before implementation, post-implementation verification protocol, stress-test step, empirical refinement, constraint quality. Start here.
compliance-trustlint.yaml — ComplyEdge TrustLint EU AI Act gate (offline, free tier). Run ./scripts/compliance/check.sh after ivd_scaffold or recipe/intent edits; CI blocks merge on critical/high violations. Recipe ships <BEGIN-COMPLYEDGE v1.0> agent block + .trustlint.yaml + pre-commit hook. No API key in repo.
canon-rules.yaml — Canon (Human Translation Layer) rules block (Phase 0a). Teaches any LLM-driven agent to emit R1 setting phase, R2 confidence markers (✓ verified / ~ inferred / ? assumed), R5 verification beats on irreversible actions, R10 folk-theory corrections, and R14 bounded identity. One canonical source → six per-client adapter views (Cursor, Cline, Claude Code, Copilot, Codex, Windsurf). Verified programmatically by the canon_check MCP tool and detected/installed via canon_check_rules_installed (never writes — always asks).

import-spec-kit.yaml — Parse a GitHub Spec Kit spec.md (User Story + Given/When/Then) into an IVD constraint scaffold via ivd_import_spec, then bind each acceptance scenario to a real, executable test — the binding Spec Kit’s own format doesn’t provide by default.
import-openspec.yaml — Parse an OpenSpec delta spec.md (Requirement + Scenario, GIVEN/WHEN/THEN/AND) via ivd_import_spec. Complements, not replaces, OpenSpec’s own /opsx:verify — that command judges coverage by LLM keyword search over the codebase; this recipe adds a requirement that a real test must pass.

coordinator-intent-propagation.yaml — Multi-agent coordination with intent delegation (coordinator writes intent for each agent it routes to)
agent-capability-propagation.yaml — Propagate agent capabilities up to coordinator routing (interface.routing); keeps LLM routing descriptions in sync with agent evolution
agent-role-based.yaml — Context-dependent agent behavior: agents adopt different personas/modes based on routing context

workflow-orchestration.yaml — Multi-step workflow orchestration across agents/modules
agent-classifier.yaml — AI classification agent pattern
self-evaluating-workflow.yaml — Continuous improvement loop: AI evaluates its own output against intent, flags gaps, re-runs

teaching-before-intent.yaml — When the user lacks technical knowledge: AI creates educational artifact, user confirms understanding, then intent flow (Principle 6 extension, canonical)
discovery-before-intent.yaml — When the user can’t describe what they want: propose goals/recipes/options, user picks, then intent flow (Principle 6 extension, experimental)

infra-background-job.yaml — Background job processing infrastructure
infra-structured-logging.yaml — Structured JSON logging for services and tool dispatch layers

data-field-mapping.yaml — Field mapping and data sources for integrations and ETL
doc-meeting-insights.yaml — Extract structured insights from meeting transcripts

Activated when a project has a .judgment/ folder at its root. See ../judgment_layer.md.

capture-correction.yaml — Capture and codify a single correction into the Judgment ledger
comparison-pair.yaml — Pair two real runs to derive diagnostic hypotheses (Rung-1 alternative to A/B)
distill-pattern.yaml — Convert an emergent pattern into a you-approvable recommendation (with optional draft recipe)

Reference the recipe in your intent artifact:

recipe: "agent-classifier"
recipe_version: "1.0"

See ../recipe-spec.md for the recipe specification and structure.

This site is open source. Improve this page.