Skip to main content
IntentLang
Draft documentation. Syntax and behavior are illustrative and will change before v1.

Ecosystem current-state map

A grounded inventory of the SkillsTech Intent Intelligence ecosystem, taken before making broad changes, so new work reuses what exists instead of forking it. This page records what is already built, where concepts are duplicated, and, most importantly, what IntentLang does and does not own.

Headline: the shared semantic backbone the vision asks for already largely exists. The risk is not missing foundations, it is duplicate graphs and overlapping engines across repos. The right work is consolidation and filling narrow gaps, not rebuilding.

Who owns what

Product Role Owns Status
IntentLang (this repo) The language + shared schema intent-graph-v1, intent-ir-v1 (superset IR), Fable rules (intent-fable-v1), the Scanner spine (intent-scan-v1), Ledger, Guardian, Simulator, Atlas, Lift, Draft, the AI-implementation contract Built
OpenThunder Verify code vs intent Verification verdicts (verification-verdict-v1), evidence packs, risk (risk-assessment-v1), SARIF-with-ship-verdict, can-i-ship; consumes il-to-ot-drift-v1 / mission-index-v1 Built
RepoMastery Deterministic comprehension / mastery Source-grounded learning: intent-learning-pack-v1, the Atlas Learning Compiler (CompiledLearningPath, teach-this-session-v1), intent-mission-mastery-v1; consumes intent-graph-v1 + mastery-pack-v1. No AI invention. Built
SkillsTech Runtime (STRU) Provider-neutral AI gateway run({ product, task, privacyMode }), provider registry (Anthropic/OpenAI-compat/Ollama/local/private), BYOK, no-silent-fallback + fail-closed privacy, audit-without-content Built
SkillsTech (monorepo) Studio + site + IDE Intent/Mission Atlas UI (app/atlas, ide/…/intentatlas), @skillstech/studio-model (graph precursor), authz/entitlements SDK Built
Skills Tech Talk AI-driven practice / training External SaaS (skillstechtalk.com); the generative pedagogy layer , asks more questions, teaches more, explain/defend/voice , calls SkillsTech Runtime for the model; consumes Repo Mastery's deterministic output External
SkillsTech Workspace Control plane Identity, entitlements, proof aggregation, provider policy Partial

The shared Intent IR already exists , do not fork it

IntentLang's intent-ir-v1 (compiler/src/intent-ir.mjs) is a strict superset of the canonical intent-graph-v1 and already carries everything the vision's node envelope asks for: id, type, title, summary, description, status, owner, source, sourceLocation, sourceType, version, hash, createdTime, updatedTime, confidence, provenance, reviewStatus, approvalStatus, permissions, sensitivity, retention, tags, evidence, metadata. It ships the exact confidence taxonomy (Confirmed / Observed / Derived / Inferred / Speculative / Conflicted), the provenance list (user-authored → human-corrected) with a factual subset, and a validateIR() honesty guard (non-factual provenance must carry a confidence and cannot be approved unless reviewed). It even declares the learning node vocabulary (LearningModule, LearningPath, Drill, Quiz, Flashcard, Misconception, LearnerConceptState).

Consequence: every product must project onto intent-graph-v1 / intent-ir-v1, never define a competing canonical graph. RepoMastery's dead architecture-ir-v1 proposal was correctly abandoned as "subsumed by intent-graph-v1."

Duplication to consolidate (the real backlog)

  1. OpenThunder ArchGraph (packages/archgraph) , a real second deterministic graph. It is reconciled up into intent-ir-v1 via archGraphToIR, and the ADRs treat IL as schema owner, but it remains the graph most at risk of drift.
  2. SkillsTech has three graph copies , @skillstech/studio-model vs app/studio/shared/* vs ide/…/graph/* (a hand-copied mirror + a TS variant). The repo's own docs flag this as the #1 consolidation risk.
  3. RepoMastery , two mastery-pack-v1 copies (consumer-lean vs producer-rich), and its learning generators overlap ~80% with SkillsTech Talk's in-app "Repository Mastery."
  4. Two IntentLang compilers , this repo's .mjs implementation (upstream owner) and SkillsTech/compiler (a TypeScript @skillstech/intentlang that ST vendors). Divergent siblings, not a file-identical fork.

Boundaries this repo will respect

  • IntentLang does not build a learning / mastery engine. RepoMastery owns the deterministic comprehension/mastery engine and already has the Atlas Learning Compiler; Skills Tech Talk owns the AI-driven practice/training on top of it. A learning compiler here would be the "second divergent mastery engine" the ecosystem constitution forbids. IntentLang's job is to emit a faithful Intent IR (intent scan --ir, or /core directly) that RepoMastery projects into lessons and Skills Tech Talk turns into practice.
  • IntentLang does not add AI SDKs or a provider router. SkillsTech Runtime (STRU) is the provider-neutral gateway with BYOK and no-silent-fallback. IntentLang's AI surface stays provider-neutral (intent draft builds a brief; STRU executes it).
  • IntentLang owns the schema and the deterministic Scanner. Filling scanner gaps (intent risks / gaps / unverified / coverage / unknowns / contradictions) and keeping the IR faithful (e.g. never-rule verification edges) is squarely in charter.

Shared identifiers and deep links

  • missionId (slug) + intentProofHash (= .intent-proof.json sourceHash) are the ratified declared↔verified join keys across IL / OT / RM / STT.
  • mission-index-v1 carries missionId + area; OpenThunder and RepoMastery join on it with no remapping.
  • Deep-link shape (RepoMastery): ?tab=atlas&mission=<missionId>&v=<intentProofHash>.

Highest technical risks

  • Graph duplication drift (ArchGraph, studio-model copies) diverging from intent-ir-v1.
  • Compiler fork drift between this repo's .mjs and SkillsTech/compiler's TypeScript.
  • Overclaiming confidential-use readiness before STRU/STW privacy controls are wired in every consumer.
  • Learning-engine overlap (RepoMastery vs SkillsTech Talk) if the shared contract is not held.

Open questions (not answerable from code)

  • Which repo becomes the single home of the IntentLang compiler (this .mjs vs ST's TS)?
  • When does SkillsTech Workspace take ownership of the ecosystem contracts from STT?
  • Is ArchGraph intended to remain OT-internal indefinitely, or collapse into intent-ir-v1?
← Ecosystem briefOne compiler, five consumers →