Skip to content

status.md

Latest commit

 

History

History
79 lines (66 loc) · 4.44 KB

status.md

File metadata and controls

79 lines (66 loc) · 4.44 KB
Raw

Project Status

A one-page summary of what is implemented, what is partially implemented, and what is explicitly not implemented. This page is the source of truth for the README's "Honest status" table — if the two ever disagree, this page wins.

Last reviewed: 2026-05-24.

Implemented

Capability Tool Notes
Read-only audit + secret scan tools/audit_workflows.py Redacts matches before write; current run reports 0 findings
Generated pack validation tools/validate_generated_pack.py 420/420 structurally valid
Unified catalog build tools/build_unified_catalog.py Deterministic quality score; not a correctness score
Catalog search tools/search_workflows.py Keyword + filter; pure stdlib
Schema validation tools/validate_schemas.py jsonschema mode if installed; lightweight fallback otherwise
Duplicate analysis tools/analyze_duplicates.py Exact + title-similarity; no auto-dedupe
Repair proposals (catalog-level) tools/propose_repairs.py Capped at 500; never auto-applied
Prompt-to-IR MVP tools/prompt_to_ir.py Deterministic keyword rules; not an LLM
IR-to-n8n exporter tools/export_ir_to_n8n.py Conservative node whitelist; placeholders only
Static n8n validation tools/validate_n8n_workflow.py Structural only; no runtime
Prompt-to-n8n orchestrator tools/prompt_to_n8n.py Writes per-prompt proof folder
Runtime repair proposals tools/propose_runtime_repair.py Schema-conformant; never auto-applied
Learning events tools/create_learning_event.py Evidence only; never auto-promoted
Human review queue tools/build_review_queue.py Aggregates everything; nothing approved
Wiki seed builder tools/build_wiki_seed.py Writes only to wiki/generated/
Public-claims audit tools/audit_public_claims.py Scans docs for overclaiming language
Final readiness check tools/final_readiness_check.py Orchestrates hard gates
CI gate set .github/workflows/audit.yml Runs audit + validators + runtime-proof smoke + review queue

Partial / MVP

Capability Why it is MVP Path to full
Prompt-to-IR Keyword-rule classifier, misses plurals and synonyms LLM-backed retrieval-grounded generator (see docs/roadmap.md)
Static n8n validator Does not parse n8n's internal parameter schemas Per-node parameter validation against an n8n type catalogue
Repair engine Only emits proposals; no apply step exists Reviewer-driven apply with explicit validationAfter recording

Not implemented

Capability Why it is not implemented
Behavioural n8n execution No n8n runtime is contacted by this repository; verifying behaviour requires a sandbox import which is not yet automated
Multi-framework export (Dify, LangGraph, Make, Zapier) Only n8n export exists; other exporters are documented but not built
LLM-backed prompt-to-workflow Outside the scope of the deterministic MVP
Autonomous self-improvement Out of scope by design; promotion to curated wiki/rules requires human PR
Production certification Not applicable — this is an open-source developer library
Regulatory compliance Not applicable — workflows are starting points; compliance is the operator's responsibility

Numbers (last measured)

These numbers come from the most recent local run. They are not historical claims and they are not marketing claims; they describe output state.

  • Workflow source files audited: ~3,613 JSON files.
  • Workflows catalogued: 3,181 entries (2,761 existing-library + 420 generated-pack).
  • possibleSecretFindings: 0.
  • invalidJsonFiles: 0.
  • Generated expansion pack: 420 structurally valid templates, no duplicate ids or slugs.
  • Unified-catalog average quality score: 86.05 (not a correctness score).
  • Catalog-level repair proposals: capped at 500 (233 high, 267 medium).
  • Runtime-proof demo cases run: 5, all PASS static validation.
  • Review queue: ~150–160 items (varies with re-runs), all status: pending.

Caveats

  • Quality score is a deterministic review-prioritisation heuristic, not a correctness or safety claim.
  • "Static validation only" applies everywhere this repository runs the validator. No claim that a workflow imports into n8n, executes correctly, or behaves as the prompt intended is made unless and until a separate behavioural validation pass records that result.
  • License: MIT. See ../LICENSE.