Amanuensis

Reference

Where to begin

Use the shortest useful reading path into an unfamiliar codebase before exploring subsystem detail.

Branch  main Checked  adc4ce04aa57 As of  2026-08-29 04:56:43 No recorded stale entries

Conspectus version: full-project survey at e33bb5f Canonical branch: main Onboarding SHA: b8b566f

§What is this codebase?

Amanuensis is an evidence-driven codebase-survey system. A host agent follows phased methodology from the skill/agent files and calls the amanuensis-memory stdio MCP server. That Node/TypeScript server owns a SQLite WAL database plus a git-backed prose storage directory and enforces survey-depth and evidence contracts. A Python subprocess renders the DB and prose into cross-linked Markdown using per-page dependency hashes. A separate Node CLI installs the agent bundle and MCP configuration. These boundaries are observed at mcp-server/src/index.ts:main@b8b566f, mcp-server/src/project.ts:resolveProject@b8b566f, and materializer/amanuensis_materializer/core.py:Materializer@b8b566f.

§Domain vocabulary

  • Conspectus: persistent evidence-bearing record of what was observed, concluded, contradicted, ruled out, or left open about a repository.
  • Disposition: terminal per-subsystem answer to one calibrated concern, with evidence.
  • Depth contract: the subsystem status bounds what claims may be written.
  • Seam: a shared boundary whose correctness cannot be inferred from either party alone.
  • Materialization: derivation of human-readable docs from durable DB/prose state.

§Directory map

directory map
PathKindCanonical source
entry-point.mdentry pointthis file
onboarding-report.mdonboarding observations and calibrationonboarding session
master-plan.mdsubsystem inventory and orderSQLite subsystem rows
findings-index.mdfinding summarySQLite findings rows
concern-checklist.mdcalibrated probesSQLite concern rows
field-notes.mdnarrative observationsfield-note rows plus prose
B-XX-*.mdsubsystem surveyone per mapped subsystem
design/delightful-output-panel/report-interface design evidenceB-06 survey
scholiast/embedded research and platform-trial evidenceB-07 survey

§Knowledge depth contract

Unmapped
None.
Scoping
File scope only.
Structural
Types, state containers, flows, concurrency; no correctness claims.
Concerns
Evidence-backed concern review and findings.
Adversarial
Concern results that survived a refutation pass.
Mapped
Packaged subsystem knowledge; seams still need composed assessment.

Any claim exceeding its source depth is speculative.

§Survey status

All seven subsystems are mapped at the current survey revision. The original five-subsystem runtime/method/materializer/delivery map remains mapped, B-06 covers report-interface design and validation, and B-07 covers the embedded research and platform-trial evidence layer. This is a coverage claim, not a defect-free claim; current finding resolution is read from durable finding records rather than this orientation paragraph.

§Open confirmed bugs

  • B07-1 (LOW): research capture and snapshot-verification calls lack a caller-controlled completion deadline (scholiast/ai-primary-web-platform-landscape/capture-component-landscape.mjs:github/npmMetadata@073cee8; scholiast/ai-primary-web-platform-landscape/verify-trial-snapshot.py:module verification@073cee8).

§Verified fixed

  • B02-1 (HIGH): the phase-gate WAL/checkpoint defect was fixed at 7bfa45c and independently verified through the durable finding-resolution gate.

§Mode selection

Bug report
Search findings, then identify its subsystem and read that survey's cited evidence. If the subsystem is below concerns depth, treat the report as an unverified hypothesis.
New feature
Use design mode: compile observed behavior, direct intent, inferred intent, constraints, contradictions, and unknowns separately before proposing options. At onboarding, read ROADMAP.md plus the target subsystem's jump-in files.
Report projection change
Read design/delightful-output-panel/design-language.md, B-06, B-04, and seams SM-06/SM-07 before changing presentation semantics.
Research-backed report change
Read B-07, the relevant scholiast/ conspectus, B-06, and seam SM-08; preserve the research claim's stated scope and limitations.
Refactor
Read the target subsystem plus every declared seam and current git validity state.
New contributor
This entry point, master-plan.md, then the target subsystem's start here.

§Minimal bootstrapping read

  1. entry-point.md
  2. master-plan.md
  3. The target subsystem's jump-in files
  4. findings-index.md
  5. concern-checklist.md for any correctness claim
  6. seam-assessments-b8b566f.md for any cross-subsystem claim