amosblomqvist
5653c77a29
Diagnosis (2026-09-11 user testing, minimal-effort GLM): sessions stuck on 'plan' forever. Two causes: (1) the three-phase view read the stored tree's currentNodeId, which nothing ever writes post author-once (2026-09-03) — a structural dead seam; (2) the teacher stochastically left quiz.nodeId untagged, and done-by-quiz-result semantics would have falsely completed nodes anyway. The redesign (docs/probe-teach-split.md) abolishes derived occupancy in favor of explicit events: - Two agents over one transcript: createProberAgent (quiz without a nodeId field, ask, probe_done) and createTeacherAgent (full toolset, plan_init onward). The chat-handler construction branch keys on a probe_done part in the transcript. MODELS.prober added. - probe_done: server-side seam tool — carries the learner assessment (strands + free-form edge read), refuses while a question is open or the probe is already concluded, and queues a handoff steer through the generalized wake pipeline (steers now dedupe by id, not jobId; job finalization only for job steers). The teacher auto-wakes and authors the plan; the learner sees the assessment via the probe-done card. - phases: probe ends at a concluded probe_done, plan ends at a successful plan_init — strictly sequential, no skipping, no legacy compatibility (data wiped at the pivot per doctrine). - Plan map = position, not completion: derivePlanStatuses reads only tagged quiz CALLS (state/outcome feed nothing); current = latest tagged node with consecutive duplicates collapsed, visited = earlier distinct stays (a revisit stays in both — current wins at render). PlanTree is structure-only; currentNodeId/doneNodeIds and the dead mutation helpers are gone. No fallback current. - Prompts split per agent (prompts/prober/, prompts/teacher/ over the trimmed shared base); graph-context injection removed entirely (shelved, docs/probe-teach-split.md §5); knowledge-graph context no longer reaches either agent. - Plan blocks are full-only snapshots at plan_init (neighborhood machinery retired; the persistent DAG window is the future UI wave); usage capture unchanged. Gates: tsc clean, 294/294 tests (phases/plan-tree/plan-blocks/question- tools rewritten to the new semantics; probe-done + handoff-wake tests added), lint, build. Live GLM smoke PASS: prober concluded with an honest assessment, teacher auto-woke, authored a 10-node plan with the block attached, then tagged checks on slope-instinct → secant derived current/ visited exactly as designed; grading honest both ways (driver off-by-one proved a wrong answer grades false).
73 lines
3.3 KiB
TypeScript
73 lines
3.3 KiB
TypeScript
import path from "node:path";
|
|
import { fileURLToPath } from "node:url";
|
|
|
|
/** Repository root (this file is at <root>/config.ts). */
|
|
export const ROOT = path.dirname(fileURLToPath(import.meta.url));
|
|
|
|
export const PATHS = {
|
|
root: ROOT,
|
|
/** Teacher system prompt — the agent's instructions (src/agent/). */
|
|
teacherPrompt: path.join(ROOT, "src", "agent", "system-prompt.md"),
|
|
data: path.join(ROOT, "data"),
|
|
/** Chat sessions (AI SDK UIMessage transcripts, one JSON file per chat). */
|
|
chats: path.join(ROOT, "data", "chats"),
|
|
graph: path.join(ROOT, "data", "graph"),
|
|
} as const;
|
|
|
|
// ─── Models ────────────────────────────────────────────────────────────────
|
|
// THE single place every LLM is chosen and tuned. .env stays for secrets
|
|
// (API keys) only — model choice and thinking budget are product tuning,
|
|
// not deployment environment, and this section is the one place they grow.
|
|
// `effort` is OpenRouter's reasoning scale (docs: reasoning-tokens):
|
|
// "none" | "minimal" | "low" | "medium" | "high" | "xhigh".
|
|
// "none" disables reasoning tokens entirely.
|
|
|
|
export type ReasoningEffort = "none" | "minimal" | "low" | "medium" | "high" | "xhigh";
|
|
|
|
export interface ModelSpec {
|
|
/** OpenRouter model id (e.g. "z-ai/glm-5.3-flash"). */
|
|
model: string;
|
|
/** Thinking budget — what the model may spend on reasoning tokens. */
|
|
effort: ReasoningEffort;
|
|
}
|
|
|
|
export const MODELS = {
|
|
/** The prober (agent A) — edge-bracketing questions; keep it lively and adaptive. */
|
|
prober: { model: "z-ai/glm-5.3-flash", effort: "minimal" },
|
|
/** The teacher (agent B) — planning + streaming, lesson-facing: smart, but keep it lively. */
|
|
teacher: { model: "z-ai/glm-5.3-flash", effort: "minimal" },
|
|
/** Background sub-agents: fast and cheap beats deep thought. */
|
|
researcher: { model: "z-ai/glm-5.3-flash", effort: "minimal" },
|
|
demoBuilder: { model: "z-ai/glm-5.3-flash", effort: "minimal" },
|
|
/** Must be vision-capable: it looks at rendered SVGs. */
|
|
visualMaker: { model: "qwen/qwen3.8-max", effort: "minimal" },
|
|
/** Extraction is judgment over given evidence — light thinking suffices. */
|
|
extractor: { model: "z-ai/glm-5.3-flash", effort: "minimal" },
|
|
/** The auditor verifies evidence exists; keep it cheap. */
|
|
auditor: { model: "z-ai/glm-5.3-flash", effort: "minimal" },
|
|
} satisfies Record<string, ModelSpec>;
|
|
|
|
export type ModelRole = keyof typeof MODELS;
|
|
|
|
/**
|
|
* The extraction pipeline (vision §1 construction): the sole graph writer.
|
|
* Triggered per chat once this many NEW quiz answers have landed since the
|
|
* chat's last extraction checkpoint. The trigger mechanism is a v1
|
|
* placeholder — deliberately isolated behind one seam
|
|
* (server/turn-service.ts) so it can be swapped without touching the
|
|
* pipeline.
|
|
*/
|
|
export const EXTRACTION = { everyNQuizAnswers: 3 } as const;
|
|
|
|
/**
|
|
* The LIVE graph store (src/store/graph.ts) — owned by the app's extraction
|
|
* pipeline. Line shapes: src/core/contracts.ts.
|
|
*/
|
|
export const LEDGER_FILES = {
|
|
nodes: path.join(PATHS.graph, "nodes.jsonl"),
|
|
edges: path.join(PATHS.graph, "edges.jsonl"),
|
|
events: path.join(PATHS.graph, "events.jsonl"),
|
|
mappings: path.join(PATHS.graph, "mappings.jsonl"),
|
|
sessions: path.join(PATHS.graph, "sessions.jsonl"),
|
|
} as const;
|