- Dressed quoted computed keys (as/satisfies/!) fail closed anywhere: the
static key survives the dressing but breaks every ['name'] matcher, and
matcher tolerance cannot span types containing ']' (as Foo['x']), so the
shape itself is the trigger (with an !(?!=) guard for ordinary
comparisons).
- Escape-built quoted keys (\u/\x/octal) fail closed: statically
resolvable, so outside the non-literal computed-member residual.
- Reflect verb indirection fails closed: any Reflect.* call naming a
write/exec verb in its argument text (Reflect.apply(db.insert, ...),
Reflect.get(db, 'insert')).
- CODE_SHAPE_REGISTER: enumerated disposition path for the fail-closed
code-shape rules (template key, dressed/escaped key, apply/call/bind,
Reflect) — a reviewed legitimate hit is registered, never resolved by
weakening the shape. Empty today; eval/new Function stays unconditional.
- KNOWN RESIDUALS: test-file/out-of-src modules named as import-graph
conduit blind spots; TYPE_ANN/DECL_LIST single-line limit documented.
- Controls E93-E102 (dressed/escaped keys at write target, conduit export,
default export, single-file factory extraction; Reflect.apply/get) plus a
clean control pinning the !== guard.
The text-only template-key rule fires on any computed member — access or
call — not only the call form, so template-keyed write targets, export
expressions, and receivers fail closed at their origin. Declaration
exports and derivations tolerate a type annotation and prior declarators
via shared TYPE_ANN/DECL_LIST shapes at all five declarator sites.
Invoking a write/exec verb through .apply/.call/.bind fails closed
anywhere (both member names statically visible, unlike the
method-extraction residual). The evasion test counts violations across
the whole synthetic chain, since some routes fail closed at the helper.
Controls E82-E92; clean control pinning the interpolated-key
discriminator.
stripExprDressing also strips trailing non-null assertions and satisfies
clauses, so export-default dressing (companies!, createDb satisfies T,
(x as unknown)!) resolves at both ends. The three ns-member right-hand
sides tolerate parenthesization. Every verb matcher admits the optional-
call form (?.() via a shared CALL_OPEN shape. A computed member call
whose key is a text-only template literal fails closed (its text never
reaches lexed code); interpolated keys stay under the non-literal
residual. Document the DB_FACTORY_IMPORTERS ns-destructure enumeration
blind spot as a residual. Controls E69-E81.
Unify member-access matching on shared DOT/BRACKET_OPEN/MEMBER_SEG shapes
(dot, bracket, ?. and ! forms) across both conduit computations, the write
targets, the factory-import clause, and the DI backstop. Treat export-default
as an expression at both ends: strip parens/assertions, classify bare-word vs
namespace-member chain, on the schema AND capability sides. Add the
capability-side export-default ns-member pass (round-5 finding 2). Admit bare
SQL qualifier words (TABLE, ONLY, IF EXISTS) between keyword and table name in
prong (ii). Controls E54-E68.
Default-export slot treated as an export name at both ends (helper-side
export-default passes, consumer-side bare-default-import resolution);
namespace-member write targets match the source's exported names via a
per-file memberSyms union; factory conduits carry a name map computed to
a fixpoint so renames propagate hop by hop (export-from, export-of-local,
export default); tracked dynamic imports of capability conduits confer
capability; factory destructure-rename from a schema namespace; static
module-namespace createRequire destructure; spread-argument writes; DCL
keywords (GRANT/REVOKE ON, LOCK TABLE); bracketed conventional DI
receiver. Permanent controls E40-E53.
Per the second M4-1a detached review (REQUEST_CHANGES, R1-R8):
- R1: prong (i) now tracks namespace destructuring, nested namespace
re-exports (ns.schema.companies), and literal dynamic-import bindings
(destructured and namespace) per contract rev 9.
- R2: capability-conduit fixpoint (computeCapabilityConduits) closes
driver/factory laundering through export-from modules.
- R3: dynamic-import check is per call site with a full-literal tail
match, so concatenated specifiers no longer pass.
- R4: computed-member calls with literal verb names (obj['insert'],
obj['query']) and spaced member access are flagged.
- R5: prong (ii) adjacency tolerates schema qualification, interposed
block comments, COPY, and escaped quotes.
- R6/R8: header documents KNOWN RESIDUALS (DI receiver rename in
capability-free files, computed non-literal member access, scan
perimeter) with the measured counterfactuals.
- createRequire fails outside a 5-module measured register (R1 route).
- 15 new permanent evasion controls (E9-E23) with helper-file conduits
run through the production analyzer.
Calibration: tree-wide prong test green with zero new exclusions;
tsc, eslint, and the full package suite pass.
Review findings F1-F8 (REQUEST_CHANGES) addressed:
- F5: replace regex comment stripping with a string-aware lexer producing
comment-free code plus accurate string/template spans (handles nesting,
regex literals, escapes)
- F3: prong (i) now tracks namespace imports and re-export conduits via a
fixpoint over the scanned import graph
- F1/F2: tagged-template exemption dropped; backlog.ts joins the infra
register; capability-gated prong (iii) — driver import flags any-receiver
execute/query/unsafe, factory import flags any-receiver execute/unsafe,
db-shaped receiver backstop covers DI'd handles; sql.raw tracked through
aliases and namespaces
- F4: closed importer enumeration added for createDb/createPgliteDb;
composition property of remaining register modules documented
- F6: plugins/ added to scan roots
- F7: per-parent slug scoping witnessed at platform_projects and workspaces
- F8: dynamic-import disposition register added; stricter-than-contract
stances documented in the header
The review's 8 evasion samples are embedded as permanent controls run
through the production analyzer, plus clean controls guarding against
false positives.
Implements docs/requirements/hierarchy-schema.md sections 2-4 and the
schema-layer half of section 6:
- Five class tables (companies, estates, platform_projects, workspaces,
hierarchy_grants) with the section 2.7 exhaustive column sets: child
node tables carry no timestamps (renames are audited via events), no
owner_id anywhere (section 4.4 - ownership is computed from grants).
- Grant constraints per section 3: exactly-one-subject and
exactly-one-target num_nonnulls CHECKs, six-column UNIQUE NULLS NOT
DISTINCT, target FKs CASCADE / principal FKs RESTRICT, six btree
indexes.
- Migration 0018 generated by drizzle-kit; SQL verified against the
contract text and applied on PGlite.
- hierarchy-schema.witness.test.ts: dual-leg witness suite (PGlite
always; real PostgreSQL under DATABASE_URL, the section 6.8 binding
leg in CI). Covers parent-FK integrity + catalog assertion, slug
scoping, column allowlist (6.2), all six grant subject/target forms,
CHECK refusals, NULLS NOT DISTINCT duplicates, NOT NULL refusals, and
deletion semantics (6.6): fail-closed parent delete, leaf cascade of
exactly its grants, principal RESTRICT.
- hierarchy-writer-coverage.test.ts: section 6.3(b) three-prong static
assertion (alias-aware symbol writes, class-table names in SQL
literals, raw-execution primitives) with empty writer allowlist,
closed infrastructure register, and closed importer enumerations for
the migration runner and migrate-tier. All prongs proven able to fire
via a planted-violation control.
Command family, audit events, and route inventory land in M4-1b.
Fresh `mosaic gateway install` (npm) left the gateway DB schema empty —
sign-in 500'd with `relation "users" does not exist`, and every entry
point (auth, bootstrap setup) failed because they all query the users
table first. Five stacked bugs on the local (PGlite) tier:
1. `packages/db/package.json` `files: ["dist"]` excluded the `drizzle/`
SQL migrations from the published tarball.
2. `runMigrations()` only supports postgres-js — unusable for embedded
PGlite.
3. `apps/gateway/src/database/database.module.ts` never invoked
migrations at startup.
4. `createPgliteDb` didn't load pgvector, so migration 0001's
`CREATE EXTENSION vector` failed.
5. Drizzle's PG migrator wraps every migration in one outer
transaction, which trips Postgres' `check_safe_enum_use` on
migration 0009 (`ALTER TYPE ADD VALUE 'pending'` → `SET DEFAULT
'pending'` in the same tx).
Changes:
- Ship `drizzle/` in the published tarball.
- `createPgliteDb` loads `@electric-sql/pglite/vector`.
- New `runPgliteMigrations(handle)` walks the Drizzle journal and
runs each statement-breakpoint chunk through PGlite's `client.exec()`
(autocommit per statement). Records into `drizzle.__drizzle_migrations`
for interop with the postgres-js path. Per-statement try/catch
surfaces which statement of which migration failed.
- `DatabaseModule` runs migrations in `OnModuleInit` before
`app.listen()`. Local tier: explicit `runPgliteMigrations` then
`storageAdapter.migrate()`. Postgres tier: just `storageAdapter.migrate()`,
which already calls `runMigrations(url)` internally — no double-call.
- Removed `packages/storage/src/test-utils/pglite-with-vector.ts`. The
"intentionally not exported" rationale is moot now that migration
0001 forces pgvector load anyway. The integration test uses
`createPgliteDb` + `runPgliteMigrations` from `@mosaicstack/db`.
Tests: BetterAuth tables exist after migrate; idempotent (re-runs 0009);
partial-failure surfaces statement-level context and leaves no ledger row.
QA on a fresh PGlite install:
- `Applying PGlite schema migrations...` then `Initializing storage
adapter (pglite)...` in startup log.
- `GET /api/bootstrap/status` → `{"needsSetup":true}` HTTP 200 (was 500).
- `POST /api/bootstrap/setup` reaches Zod validator (was 500).
Scope: this PR fixes the local (PGlite) tier. Postgres-tier first
install still has the outer-transaction problem and a journal ordering
bug (0009's `when` < 0008's). Documented inline as TODO and in the
scratchpad — needs a separate change with real-Postgres validation.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
Publish pipeline:
- Add publish-npm step to .woodpecker/publish.yml — publishes all
@mosaic/* packages to Gitea npm registry on main push/tag
- Requires gitea_npm_token Woodpecker secret (package:write scope)
- publish-npm runs after build, parallel with Docker image builds
- pnpm publish resolves workspace:* to concrete versions automatically
Package configuration:
- All 20 packages versioned at 0.0.1-alpha.1
- publishConfig added to all packages (Gitea registry, public access)
- files field added to all packages (ship only dist/)
- @mosaic/forge includes pipeline/ assets in published package
Meta package (@mosaic/mosaic):
- Now depends on @mosaic/forge, @mosaic/macp, @mosaic/prdy,
@mosaic/quality-rails, @mosaic/types
- npm install @mosaic/mosaic pulls in the standalone framework
Build fixes:
- Fix forge and macp tsconfig rootDir: '.' -> 'src' so dist/index.js
resolves correctly (was dist/src/index.js)
- Exclude __tests__ and vitest.config from build includes
- Clean stale build artifacts from old rootDir config
Required Woodpecker secret:
woodpecker secret add mosaic/mosaic-stack \
--name gitea_npm_token --value '<token>' \
--event push,manual,tag
The insights table uses vector(1536) but no migration enables the pgvector
extension. CI postgres (pgvector/pgvector:pg17) has the extension available
but it must be explicitly created before use.
Adds CREATE EXTENSION IF NOT EXISTS vector at the top of
0001_cynical_ultimatum.sql (the first migration referencing vector type).
The migration file 0001_cynical_ultimatum.sql existed on disk but was not
registered in the Drizzle journal (_journal.json). This caused fresh-database
migrations (CI) to skip creating tables (agent_logs, insights, preferences,
skills, summarization_jobs), then 0002_nebulous_mimic.sql would fail trying
to ALTER the non-existent preferences table.
Fix: insert cynical_ultimatum at idx 1 in the journal and shift all
subsequent entries (idx 2-7).
Verified: pnpm test passes (347 tests, 35 tasks).
- DB client: configure connection pool (max=20, idle_timeout=30s, connect_timeout=5s)
- DB schema: add missing indexes for auth sessions, accounts, conversations, agent_logs
- DB schema: promote preferences(user_id,key) to UNIQUE index for ON CONFLICT upsert
- Drizzle migration: 0003_p8003_perf_indexes.sql
- preferences.service: replace 2-query SELECT+INSERT/UPDATE with single-round-trip upsert
- conversations repo: add ORDER BY + LIMIT to findAll (200) and findMessages (500)
- session-gc.service: make onModuleInit fire-and-forget (removes cold-start TTFB block)
- next.config.ts: enable compress, productionBrowserSourceMaps:false, image avif/webp
- docs/PERFORMANCE.md: full profiling report and change impact notes
fred
code-infra-01andorch-01
ops-deploy-01andorch-01
jason.woltje
jarvis