Implement the full ADR-0006 plan: three-layer separation of transaction
correctness, retry policy, and cross-process writer coordination.
Layer 1 — scripts/tx.ts (transaction correctness):
- runWriteTransaction executes work exactly once; no internal retry.
- BEGIN IMMEDIATE takes the write lock up front (avoids SQLITE_BUSY_SNAPSHOT).
- Guarded rollback: checks inTransaction() via adapter before attempting
ROLLBACK; never masks the primary exception.
- WriteTxDiagnostics attached to errors: phase, code, label,
rollbackSucceeded, rollbackError, transactionActive.
- Binding adapters (betterSqliteTransactionAdapter, nodeSqliteTransactionAdapter)
mapping better-sqlite3's `.inTransaction` and node:sqlite's `.isTransaction`.
- configureConnection centralizes WAL + synchronous + busy_timeout.
Layer 2 — scripts/write-coordinator.ts (retry policy):
- runRetryableWriteTransaction: bounded retry with total time budget.
- Only retries when the transaction confirmed ended (transactionActive=false)
and the error is SQLITE_BUSY during work/commit phase.
- BEGIN-phase BUSY = abort entire build (isBeginBusyFailure); the caller
returns `{ deferred: true, reason: 'writer_busy' }` instead of waiting.
- hasUnusableTransaction detects a still-active transaction after failure;
aborts the build immediately, never retries.
Layer 3 — scripts/writer-lease.ts (cross-process coordination):
- acquireWriterLease: dedicated writer.lock.sqlite with busy_timeout=0 +
BEGIN IMMEDIATE. Non-blocking attempt; bounded wait with retryDelayMs.
- writerLockPathFor derives lock path from the target DB path.
- Lease held for the entire build; released on completion or failure.
- Lock DB uses DELETE journal (not WAL); crash/close auto-releases.
- All consumers obey: skill acquires at build start (returns deferred if
unavailable); app daemon (via worker) acquires for its build cycle.
Build semantics changes:
- affectedSessionIds updated only after successful commit.
- BuildIndexResult gains skipped/skippedFiles for observability.
- Skill finalize failure now fails the build (was silently warned).
- Checkpoint changed to PASSIVE (TRUNCATE reserved for maintenance/exit).
- Skill buildIndex returns { deferred, reason } on lease contention;
indexer-service reschedules the build (deferredRetryMs) without publishing
a heartbeat (so the build-deferred state is visible to cross-process
arbitration).
- Service publishes heartbeat immediately on start() for correct arbitration.
Tests:
- tests/write-transaction.test.mjs: single-shot execution, diagnostics
propagation, auto-rolled-back transaction detected, rollback failure
captured as metadata, BEGIN IMMEDIATE semantics.
- tests/writer-lease.test.mjs: acquire/release, contention returns null,
bounded wait with release during budget.
- tests/app-writer-lease.test.mjs: better-sqlite3 adapter integration.
- tests/app-rollback-guard.test.mjs: rewritten — transient BUSY recovered
by coordinator, persistent BUSY skips file, begin-busy aborts build,
live-transaction aborts build, phantom affectedSessionIds prevented.
- tests/daemon-arbitration.test.mjs: skill defers to fresh app heartbeat,
builds when heartbeat is stale.
- tests/app-indexer-service.test.mjs: new cases for deferred-retry
scheduling and immediate heartbeat on start.
- app/tests/electron-concurrency.mjs + child: dual-child IPC structure for
real better-sqlite3 contention (holder acquires lock → build child starts
→ delayed release → result collected; persistent contention bounded).
ADR-0006 updated to reflect the implemented design.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
95 lines
4.1 KiB
Markdown
95 lines
4.1 KiB
Markdown
# Obelisk
|
|
|
|
Obelisk is explicit memory infrastructure for coding agents: it indexes local
|
|
Claude Code and Codex transcripts into a queryable SQLite evidence layer, and a
|
|
CodeAct runtime lets an agent write a small query, run it, and answer from real
|
|
session history. This glossary pins the terms that are specific to Obelisk; it is
|
|
not a spec.
|
|
|
|
## Runtime interface
|
|
|
|
**Runtime interface**:
|
|
The public contract, expressed as four verbs — `build`, `search(text)`,
|
|
`query(code)`, `attune(code)`. Skill, CLI, and MCP are transports over this same
|
|
shape; none of them add their own retrieval surface.
|
|
_Avoid_: API, tool surface
|
|
|
|
**CodeAct**:
|
|
The interaction style where an agent writes JavaScript that runs inside the
|
|
`query(code)` sandbox and returns JSON, rather than calling many fine-grained
|
|
tools. This is Obelisk's core design choice.
|
|
_Avoid_: tool-calling, function-calling
|
|
|
|
**Helper**:
|
|
A convenience accessor available only inside the `query(code)` sandbox
|
|
(`overview`, `search`, `context`, `sql`, `memories`, …). Helpers are never
|
|
promoted to an external tool surface.
|
|
|
|
## Indexing
|
|
|
|
**Provider adapter**:
|
|
A pure per-source module (claude, codex, later opencode, pi, …) that discovers a
|
|
source's transcript files and parses one into a stream of records. It never opens
|
|
or writes a database; adding a source means adding one adapter. The shared pure
|
|
parse/discover helpers live in `scripts/parsing.mjs`, which imports only
|
|
node:fs/path/os — deliberately node:sqlite-free so the compiled providers can be
|
|
consumed by the app (whose Electron runtime has no `node:sqlite`).
|
|
_Avoid_: parse core, parser, ingest
|
|
|
|
**Record**:
|
|
One normalized row destined for the index (session, message, tool call, tool
|
|
result, summary, subagent, workflow, …), emitted by a provider adapter before any
|
|
persistence happens.
|
|
|
|
**Persist layer**:
|
|
The single shared, provider- and binding-agnostic writer that consumes records
|
|
from any adapter and writes them into an injected SQLite handle inside a
|
|
transaction. The binding is injected — `node:sqlite` (skill/CLI) or
|
|
`better-sqlite3` (app) — so there is one persist implementation, not one per
|
|
binding.
|
|
_Avoid_: writer, sink, DAO
|
|
|
|
**Daemon indexing mode**:
|
|
Continuous incremental indexing driven by a long-lived process (the desktop app,
|
|
later a CLI daemon) that watches transcript directories and keeps the index fresh
|
|
as files change.
|
|
_Avoid_: watcher mode, live indexing
|
|
|
|
**Passive pull mode**:
|
|
On-demand incremental indexing performed by the skill when there is no active
|
|
daemon: an invocation of the runtime brings the index up to date, then answers.
|
|
_Avoid_: lazy indexing, on-read indexing
|
|
|
|
**index_state**:
|
|
The bookkeeping table shared by both indexing modes. It records, per transcript
|
|
path, the last-seen `mtime` and `lines_processed` (enabling resume-from-line
|
|
incremental indexing), plus heartbeat/last-build markers used for daemon
|
|
arbitration.
|
|
|
|
**Daemon arbitration**:
|
|
The policy by which the passive pull mode detects a fresh daemon from the
|
|
`__app_heartbeat__` marker and skips every skill-side mutation, including schema
|
|
setup, indexing, checkpointing, and `attune`. The heartbeat alone means “the
|
|
daemon should write”; `__app_last_successful_build__` records coverage/freshness,
|
|
not ownership. Both indexing modes use the same persist layer.
|
|
|
|
**Writer lease**:
|
|
The hard cross-process safety mutex behind daemon arbitration. A writer holds
|
|
`BEGIN IMMEDIATE` on `.obelisk/writer.lock.sqlite` for the complete mutation;
|
|
manual rebuild holds it through build, target-database replacement, and reopen.
|
|
The heartbeat expresses policy, while the writer lease prevents overlapping
|
|
writes during races, stale heartbeats, or processes from different versions.
|
|
|
|
## Memory
|
|
|
|
**Queryable session memory**:
|
|
The evidence layer — real sessions, messages, tool calls, subagents, workflows —
|
|
that an agent queries on demand. Obelisk deliberately does this instead of
|
|
implicit/ambient memory.
|
|
_Avoid_: implicit memory, ambient memory, auto-recall
|
|
|
|
**Approved durable memory**:
|
|
Human-approved conclusions persisted as markdown plus a registry record, via
|
|
`attune(code)` calling `remember()`/`forget()`. Auditable and revocable.
|
|
_Avoid_: long-term memory, vector memory
|