Docs · Référence

Catalogue des règles.

Les 64 règles de craft que la porte peut exiger, chacune reliée à une ou plusieurs phases et à un risque OWASP ASI. Table générée depuis le contrat machine stable runward rules --json.

64 règles · runward v0.22.0

Règle	Impact	Phases	ASI	Pourquoi
Day-Zero Project Setup Checklistchecklist-day-zero-project	CRITICAL	advisory	—	Starts a project on the foundation that prevents costly refactors, with the core engineering disciplines wired from the first commit
Pre-Production Resilience Checklistchecklist-pre-production-resilience	CRITICAL	advisory	ASI08	Ensures application handles failures gracefully in production
Pre-Production Security Checklistchecklist-pre-production-security	CRITICAL	advisory	ASI02, ASI03, ASI06	Ensures security requirements are met before production deployment
Secrets at the Network Boundary, Never in the Modelconfig-secrets-boundary	CRITICAL	floor, govern	ASI03, ASI10	Makes secret disclosure structurally impossible by keeping the real key out of the model and the domain
Contract Governance (Versioned, Additive, Expand-then-Contract)contracts-governance	CRITICAL	architect	ASI04, ASI07	Lets the system evolve without breaking consumers, by governing the contract at the boundary rather than the implementation behind it
Evaluation Loop (Test + Evaluate, Hold-out, Anchored Judge)eval-loop	CRITICAL	govern	ASI08	Gives a non-deterministic system a trustworthy way to detect behavioural regressions without letting it optimise its own grader
Deterministic Boundary of the Modelfrontier-deterministic-boundary	CRITICAL	floor	ASI01	Keeps every fact, figure and decision that can be checked out of the model, so the system is verifiable and cannot hallucinate load-bearing values
The Redone-Task Proofhandover-redone-task-proof	CRITICAL	handover	ASI09	A hand-over is proven by a real task redone end to end without the departing builder — recorded, dated, pointed at — so autonomy is a fact, not a farewell promise
Move the Deterministic out of the Modelhexa-move-deterministic-out	CRITICAL	floor	ASI01, ASI02	Reduces LLM costs and latency by moving deterministic logic out of LLM calls
Model Provider as an Adapter (Detected, Never Hardcoded)provider-llm-auto-detection	CRITICAL	floor	—	Keeps the model behind a port so the provider can change by configuration without touching the core
Fail-Open for the Non-Critical, Fail-Closed for the Sensitiveresilience-fail-open	CRITICAL	govern	ASI02, ASI08	Degrades gracefully where safe and denies safely where it matters, instead of applying one failure policy everywhere
If Doubt, Upgrade the Modelrouting-confidence-upgrade	CRITICAL	advisory	—	Prevents quality degradation from incorrect model selection
Externalize State for Scalingscaling-state-externalization	CRITICAL	advisory	—	Enables horizontal scaling by removing in-process state dependencies
Unexpected Code Execution Runs Sandboxed, Never In-Processsecurity-code-execution-sandbox	CRITICAL	govern	ASI05, ASI02	Caps the blast radius of any code the agent runs (generated or tool-provided) by isolating it from the host, secrets and network
Pin Every Consumed Tool/MCP Server by Version and Hashsecurity-mcp-server-pinning	CRITICAL	architect, govern	ASI04, ASI10	A consumed tool or MCP server is pinned by version and content hash, so a supply-chain swap or rug-pull cannot silently change what the agent may call
Untrusted Input, Prompt Injection and the Lethal Trifectasecurity-prompt-injection	CRITICAL	floor, govern	ASI01, ASI06	Treats prompt injection as a first-rank, structural threat constrained by architecture rather than detected by heuristics
Sovereignty Graduated by Class of Datatopology-sovereignty-by-data-class	CRITICAL	topology	—	Prevents regulated or personal data from silently landing on a placement that cannot hold its sovereignty
Background Job Guardrails (Retry, Idempotency, Bounded Concurrency, Job Observability)async-job-guardrails	HIGH	govern	ASI08	The four non-negotiables of any background job — bounded retry per step, idempotency under concurrency, capped and partitioned concurrency, and queue lag plus failure rate as first-class metrics
Post-Turn Pipeline (Keep the Interactive Turn Thin)async-post-turn-pipeline	HIGH	advisory	—	Keeps response latency bound to what the answer needs by moving extraction, consolidation and summarization into an isolated, event-driven pipeline that runs after the turn
Pre-Production Observability Checklistchecklist-pre-production-observability	HIGH	advisory	—	Ensures you can monitor and debug production issues
Pre-Production Performance Checklistchecklist-pre-production-performance	HIGH	advisory	—	Ensures application performs well under production load
Memory Carries Provenance; an Untrusted Write Cannot Act Unrevieweddata-memory-provenance	HIGH	govern	ASI06	Every stored memory item carries its origin and trust tier; a memory written from untrusted content is quarantined and cannot influence a privileged action without human review
Forward-Only, Branch-First Migrationsdata-migrations-forward-only	HIGH	advisory	—	Keeps schema change safe and auditable by making migrations additive, ordered, and tested on a branch before main
The Agent Charter Is Finalized as the Leave-Behindhandover-agents-charter-final	HIGH	handover	ASI10	AGENTS.md leaves the mission as the standing constraint on every future agent — verification commands, judgment boundaries, never/PR rules — so the next agent inherits the discipline, not just the code
The Runbook Is Executable, Not Descriptivehandover-runbook-executable	HIGH	handover	ASI08	The recovery runbook carries real commands for the gestures that matter at 3 a.m. — start, observe, debug, resume, swap the provider, rerun the bench, process a suspended approval — so an incident meets procedures, not prose
The Succession Is Namedhandover-succession-named	HIGH	handover	—	The system leaves the mission with a named owner, an escalation path and a review cadence — an unowned agentic system in production is an incident with a start date
New Feature = New Adapterhexa-adapter-pattern	HIGH	architect, floor	—	Keeps core business logic clean and external integrations isolated
Hexagonal Architecture Structurehexa-architecture	HIGH	architect, floor	—	Enables testability, maintainability, and clean dependency management
A Thin Model Abstraction You Own (Not a Heavy Chain Framework)hexa-typescript-native	HIGH	architect	—	Reduces token overhead, improves debuggability, removes volatile dependencies by keeping a light abstraction you control instead of a heavy framework
Tiered Memory Routerpatterns-memory-router-tiered	HIGH	advisory	ASI06	Order-of-magnitude latency cut on simple queries by fetching only the memory tier the query needs, measured on a reference system — recalibrate on your traffic
ADR Discipline and Working Journalprocess-adr-and-journal	HIGH	architect	—	Makes structural decisions traceable and reversible-by-record, so the system can be picked up by anyone without re-litigating settled choices
Graceful Degradation of Optional Servicesprovider-no-crash-missing-env	HIGH	floor	—	Lets the app start and run wherever it is deployed, enabling features by available config rather than crashing on a missing variable
Multi-Provider Fallbackresilience-multi-provider-fallback	HIGH	govern	ASI04, ASI08	Ensures LLM availability by falling back to alternative providers
LLM Retry with Exponential Backoffresilience-retry-backoff	HIGH	govern	ASI08	Handles transient LLM failures gracefully without overwhelming the API
Model Cost Ratiosrouting-model-cost-ratios	HIGH	advisory	—	Essential reference for cost optimization decisions
Smart Routing by Complexityrouting-smart-complexity	HIGH	advisory	—	Order-of-magnitude latency cut on simple requests and a substantial cost reduction through intelligent model selection, measured on a reference system — recalibrate on your traffic
Distributed Rate Limitingscaling-distributed-rate-limiting	HIGH	advisory	ASI08, ASI10	Enables consistent rate limiting across multiple instances
Provenance on Every Field, So the Human Never Trusts a Guesssecurity-human-agent-trust	HIGH	govern	ASI09	Stops a model-proposed value being read as a verified fact at the approval point — the human decides on real provenance, never on fabricated confidence
Re-approve a Tool on Any Change to Its Signed Definitionsecurity-tool-change-reapproval	HIGH	govern	ASI02, ASI04	A change to a registered tool's definition (name, schema, description, permissions) forces re-approval before the agent may call it, closing the rug-pull window
State as an Immutable Journal (Event Sourcing, Replay, Provenance)state-event-sourcing	HIGH	floor	ASI06	Makes the agent a stateless reducer over an external, replayable journal, so truth, audit and recovery are structural rather than bolted on
Tool Registry Patterntools-registry-pattern	HIGH	advisory	ASI02, ASI03	Replaces giant switch statements with maintainable, extensible tool management
Tool Scope and Atomicitytools-scope-atomicity	HIGH	floor	ASI02	Reduces token usage and improves LLM tool selection accuracy
Every Port Has a Named Placementtopology-port-placement-mapped	HIGH	topology	—	Keeps the execution topology traced port by port, so a placement never drifts unseen behind the domain
Trace Export Is a Decision, Because Traces Are Datatopology-trace-export-decision	HIGH	topology	—	Stops the most sensitive payloads leaving under the name "telemetry" without a named, reviewed decision
A Usage Registry, Because Risk Is Classed by Deploymenttopology-usage-registry-present	HIGH	topology	—	Lets you answer, at audit time, what runs where touching what — risk attaches to a deployment, not a platform
Scheduled Maintenance (The Cron Updates and Cools, It Never Destroys Truth)async-scheduled-maintenance	MEDIUM	advisory	—	Keeps memory, derived views and the journal healthy over months of operation through periodic decay, pruning and cold archiving that never touch the source of truth
Prompt Cache Stability (a Cost Lever, Provider-Specific)cache-three-tier-architecture	MEDIUM	advisory	—	Cuts input-token cost when a provider bills cached prefixes cheaper, by keeping the stable part of the prompt byte-identical across requests
Config Typing with Zodconfig-typing-zod	MEDIUM	advisory	—	Makes configuration type-safe and validated once at the boundary, so the rest of the system trusts it
Reversible Memory Consolidation (Merge, Summarize, Keep Pointers)data-memory-consolidation	MEDIUM	advisory	—	Keeps long-term memory compact without destroying evidence, so every consolidated fact can be unfolded back to its sources at audit time
Bi-Temporal Memory Invalidation (Contradicted Facts Are Dated, Not Deleted)data-memory-invalidation	MEDIUM	advisory	—	Lets memory change its mind without lying about the past, by tracking when a fact was learned and when it was valid as two separate timelines
Memory Scoring Formuladata-memory-scoring	MEDIUM	advisory	—	Provides optimal memory retrieval ranking
Orphan Data Cleanupdata-orphan-cleanup	MEDIUM	advisory	—	Prevents data bloat from orphaned records
TTL by Data Typedata-ttl-types	MEDIUM	advisory	—	Ensures proper data lifecycle management and compliance
Stack Posture (Sober Default plus Trigger), Not a Fixed Stackhexa-recommended-stack	MEDIUM	advisory	—	Replaces a frozen product stack with a portable rule for choosing technology by evidence, so the rule does not age with any vendor
Alert Configurationobservability-alert-configuration	MEDIUM	advisory	—	Ensures timely notification of production issues
LLM Metrics Trackingobservability-llm-metrics	MEDIUM	advisory	—	Enables cost monitoring, performance optimization, and anomaly detection
Structured JSON Logsobservability-structured-json-logs	MEDIUM	advisory	—	Enables log querying, aggregation, and alerting in production
Prompt as Program (Prompt Compiler)patterns-prompt-compiler	MEDIUM	advisory	—	Enables versioning, testing, and optimization of prompts
Request ID Propagationpatterns-request-id-propagation	MEDIUM	advisory	—	Enables end-to-end request tracing across services
Codebase Metrics Thresholdsquality-codebase-metrics	MEDIUM	advisory	—	Identifies code quality issues before they become problems
Zod Input Validationquality-zod-input-validation	MEDIUM	advisory	—	Prevents invalid data from entering the system and provides type safety
Retryable vs Non-Retryable Errorsresilience-retryable-errors	MEDIUM	advisory	—	Prevents wasted retries and ensures quick failure for unrecoverable errors
Database Connection Poolingscaling-db-connection-pooling	MEDIUM	advisory	—	Prevents connection exhaustion under load
Log Providers at Startupobservability-startup-provider-log	LOW	advisory	—	Simplifies debugging by showing configuration at startup

Généré depuis runward rules --json — exécutez runward explain <slug> pour le texte complet · runward v0.22.0

← Docs