@happyvertical/smrt-cli 0.60.9 → 0.60.11

package/AGENTS.md

@@ -95,14 +95,8 @@ it in the same migration also run outside the transaction, in order.
   resumes at the index build — their `error_message` carries a
   `[smrt: concurrent-index phase 1 committed]` marker, so the non-index
   statements that already committed are not re-run.
-- INVALID indexes — the stump a cancelled or timed-out
-  `CREATE INDEX CONCURRENTLY` leaves, which `pg_indexes` still reports as
-  present — are detected via `pg_index.indisvalid` and dropped before the
-  rebuild.
-- `migrations.postgres.useConcurrently: false` vetoes the flag; index DDL then
-  runs inside the atomic transaction (still bounded by the timeouts).
-- Without the flag, a batch containing explicit `CONCURRENTLY` DDL is rejected
-  before the transaction opens — PostgreSQL cannot run it there.
+- INVALID-index cleanup, the `useConcurrently: false` veto, and the explicit-`CONCURRENTLY` rejection:
+  agents/db-migrate-concurrency.md.
 
 ## `doctor`'s agent-surface section (#2591)
 
@@ -283,51 +277,8 @@ exactly those five hardcoded names and nothing else — **never** derived from
 drop `includeDroppedTables` (default `false`, both call sites explicit —
 `differ.ts:450`, `db-diff.ts:210`) deliberately refuses to be.
 
-- Execute-or-refuse, like `db:rollback`: it refuses (non-zero exit, nothing
-  dropped) if any target table has rows, has anything but that table's own
-  expected columns and types, or is referenced by a foreign key anywhere in
-  the live database. A consumer's own unrelated table sharing one of these
-  five names is exactly what the shape check protects.
-- The expected shape is **per table, not one universal baseline**:
-  `smrt_objects`/`smrt_classes`/`smrt_collections` are exactly `id`/`slug`/
-  `context`/`created_at`/`updated_at`, but `smrt_hierarchicals` also
-  requires `parent_id` and `smrt_polymorphic_associations` also requires
-  `meta_type`/`meta_id`/`role`/`sort_order` — `SmrtHierarchical`'s and
-  `SmrtPolymorphicAssociation`'s own real fields, confirmed against a
-  genuine pre-#2644 `db:migrate` run. A table under either of those two
-  names with *only* the plain five columns is refused as unexpected shape,
-  not treated as safe.
-- Companion indexes are enumerated from each table's live schema (never
-  guessed). `smrt_hierarchicals`/`smrt_polymorphic_associations` do not
-  carry the `created_at` list-ordering index the other three do (also
-  empirically confirmed, not assumed).
-- Bounded PostgreSQL execution reuses the same `SET LOCAL lock_timeout` /
-  `statement_timeout` pattern as every other DDL path (#2362). Every target
-  is locked `IN ACCESS EXCLUSIVE MODE` before re-checking it — a plain
-  `SELECT COUNT(*)` alone would not block a concurrent writer — and shape,
-  type, and emptiness are all re-verified inside that transaction
-  immediately before dropping anything, via raw `information_schema.columns`
-  / `PRAGMA table_info` (`getTableSchema()` is not available on the
-  transaction-scoped connection). Only `DROP TABLE` is ever executed —
-  never a `DROP INDEX` by name — so `DROP TABLE`'s own cascade removes the
-  table's indexes identity-safely instead of trusting a possibly-stale
-  planned index name.
-- `--dry-run` prints the exact plan (which tables exist, their companion
-  indexes, and the DROP statements) without executing.
-- Skipping it is safe: a table left behind is inert and permanently orphaned,
-  never written to or read from again. Documented, narrow residual risk, all
-  schema-integrity (never data loss — the target is always verified empty
-  first): DuckDB's adapter does not report foreign keys through
-  `getTableSchema()` at all; this module's reverse-FK scan (like every other
-  PostgreSQL schema tool in this package) only looks at the `public` schema;
-  and foreign keys are checked at planning time only, not re-scanned inside
-  the execution transaction. On PostgreSQL a foreign key that appeared after
-  planning is still caught there — `DROP TABLE` refuses when a real
-  dependent exists, dependency-based, empty parent or not. **This does not
-  hold on SQLite**: its FK enforcement on `DROP TABLE` is row-based, so an
-  empty parent (this command's precondition) drops cleanly past a real
-  foreign key, verified directly. The plan-time scan is the only gate on
-  SQLite/DuckDB, narrower than PostgreSQL's.
+Execute-or-refuse, per-table shape checks, lock/timeout behavior, `--dry-run`, and the residual risks of
+skipping it: agents/db-drop-framework-base-tables.md.
 
 ## Architecture
 
@@ -344,6 +295,26 @@ drop `includeDroppedTables` (default `false`, both call sites explicit —
 - `src/loaders/` — class-loader, local-loader, npm-loader, git-loader, template-loader
 - `src/discovery/manifest-discovery.ts` — manifest auto-discovery
 - `src/commands/docs-claude.ts` — downstream AGENTS.md generation + Claude alias; no object scanning
+- `src/commands/docs-knowledge-bundle.ts` — knowledge-bundle half of `docs:agents`
+- `src/commands/docs-skills-install.ts` — skill-installation half of `docs:agents` (#4708)
+
+## docs:agents knowledge bundle (consumer contract)
+
+Beside the output file, `docs:agents` writes `smrt-knowledge/` (`.agents/smrt-knowledge/`; `docs:claude` → `.claude/`):
+`bundle.json` (`smrt-okf-bundle/1`: `exportedAt` kept while content is unchanged, `packages[{name,version,concept}]`,
+`intents`), `components/<unscoped>.md` per installed package shipping `dist/smrt-knowledge.okf.md` with
+`package_version` re-stamped from the installed `package.json`, and `intents/*` limited to intents with ≥1 installed
+home. `smrt-framework.md` leads with `## Find it by intent` over those same intents. Pages come from the published
+artifacts — `scripts/ship-knowledge-pages.mjs` (root `pnpm build`, `prepack-package.js`) copies the committed
+`docs/knowledge` pages into `dist/`; intents ship in smrt-cli `dist/knowledge/intents/` — never from a checkout (monorepo
+runs fall back to `docs/knowledge`). Links to uninstalled packages' pages dangle by design. The shape lives in
+`@happyvertical/smrt-dev-mcp/knowledge` (`okf-bundle.ts`); `dev:knowledge-check` reads it (`stale-okf-bundle`).
+
+Skill install (#4708): unless `--no-install-skills`, `docs:agents` also writes the `smrt-catalog` skill from
+`@happyvertical/smrt-dev-mcp` to `.claude/skills/smrt-catalog/SKILL.md` and `.agents/skills/smrt-catalog/SKILL.md`
+under the working directory (skipped on `--dry-run`). The copy carries a `<!-- smrt-generated: … smrt-dev-mcp@<version> … -->`
+marker after the frontmatter: a rerun replaces a marked copy, a copy without the marker is left alone with a warning.
+The shipped text names `.agents/smrt-knowledge/`; the installer substitutes the real bundle dir (`docs:claude`, custom `--output`). Only the smrt repository itself (root `package.json` name `@happyvertical/smrt`, not any pnpm monorepo) also gets `smrt-contribute`; an unmarked copy byte-identical to the shipped skill (the repo's own `.agents/skills/smrt-contribute`) counts as current. Destinations go through `containedPath` and are refused when any path segment is a symlink, before any write.
 
 ## Gotchas
 

package/README.md

@@ -186,6 +186,11 @@ alias supports the same flag and retains its historical output path.
 Regenerate snapshots after moving the project or updating dependencies so the
 local source paths remain current.
 
+`docs:agents` also installs the `smrt-catalog` agent skill into
+`.claude/skills/` and `.agents/skills/` so an agent consults the knowledge
+catalog before hand-rolling infrastructure. Pass `--no-install-skills` to skip
+it. A copy you edited (the generated marker removed) is never overwritten.
+
 Review and architecture prompt bundles include package guidance and modules
 matched by file or focus hints. Add `--complete` to either context command to
 include every module for the selected packages. These commands still build the

package/agents/app-commands.md

@@ -97,6 +97,8 @@ config loading; `bin/smrt.js` imports `dist/` in-process so SIGTERM reaches
   publishes via `link()` (never replaces); import requires a 0600 regular
   bundle for the same app, an exactly matching manifest schema, and an empty
   target; deployed import requires `SMRT_MAINTENANCE_MODE=true`.
+  Local relative asset locators resolve against the approved asset root during
+  export and stay relative after import. Legacy file URLs import as file URLs.
 - **Process identity**: `app.pid` is trusted only when the live command
   line's last two tokens are a path with basename `smrt-web.mjs` and exactly
   `--smrt-instance=<record instance>` (an app started by the template's former

package/agents/db-drop-framework-base-tables.md

@@ -0,0 +1,49 @@
+# `db:drop-framework-base-tables` details (#2647)
+
+Detail for the one-time remediation described in `AGENTS.md`.
+
+- Execute-or-refuse, like `db:rollback`: it refuses (non-zero exit, nothing
+  dropped) if any target table has rows, has anything but that table's own
+  expected columns and types, or is referenced by a foreign key anywhere in
+  the live database. A consumer's own unrelated table sharing one of these
+  five names is exactly what the shape check protects.
+- The expected shape is **per table, not one universal baseline**:
+  `smrt_objects`/`smrt_classes`/`smrt_collections` are exactly `id`/`slug`/
+  `context`/`created_at`/`updated_at`, but `smrt_hierarchicals` also
+  requires `parent_id` and `smrt_polymorphic_associations` also requires
+  `meta_type`/`meta_id`/`role`/`sort_order` — `SmrtHierarchical`'s and
+  `SmrtPolymorphicAssociation`'s own real fields, confirmed against a
+  genuine pre-#2644 `db:migrate` run. A table under either of those two
+  names with *only* the plain five columns is refused as unexpected shape,
+  not treated as safe.
+- Companion indexes are enumerated from each table's live schema (never
+  guessed). `smrt_hierarchicals`/`smrt_polymorphic_associations` do not
+  carry the `created_at` list-ordering index the other three do (also
+  empirically confirmed, not assumed).
+- Bounded PostgreSQL execution reuses the same `SET LOCAL lock_timeout` /
+  `statement_timeout` pattern as every other DDL path (#2362). Every target
+  is locked `IN ACCESS EXCLUSIVE MODE` before re-checking it — a plain
+  `SELECT COUNT(*)` alone would not block a concurrent writer — and shape,
+  type, and emptiness are all re-verified inside that transaction
+  immediately before dropping anything, via raw `information_schema.columns`
+  / `PRAGMA table_info` (`getTableSchema()` is not available on the
+  transaction-scoped connection). Only `DROP TABLE` is ever executed —
+  never a `DROP INDEX` by name — so `DROP TABLE`'s own cascade removes the
+  table's indexes identity-safely instead of trusting a possibly-stale
+  planned index name.
+- `--dry-run` prints the exact plan (which tables exist, their companion
+  indexes, and the DROP statements) without executing.
+- Skipping it is safe: a table left behind is inert and permanently orphaned,
+  never written to or read from again. Documented, narrow residual risk, all
+  schema-integrity (never data loss — the target is always verified empty
+  first): DuckDB's adapter does not report foreign keys through
+  `getTableSchema()` at all; this module's reverse-FK scan (like every other
+  PostgreSQL schema tool in this package) only looks at the `public` schema;
+  and foreign keys are checked at planning time only, not re-scanned inside
+  the execution transaction. On PostgreSQL a foreign key that appeared after
+  planning is still caught there — `DROP TABLE` refuses when a real
+  dependent exists, dependency-based, empty parent or not. **This does not
+  hold on SQLite**: its FK enforcement on `DROP TABLE` is row-based, so an
+  empty parent (this command's precondition) drops cleanly past a real
+  foreign key, verified directly. The plan-time scan is the only gate on
+  SQLite/DuckDB, narrower than PostgreSQL's.

package/agents/db-migrate-concurrency.md

@@ -53,3 +53,14 @@ are written after the diff, when both runs have already planned.
   — two concurrent CLI runs apply one batch; a held lock blocks a run before it
   touches schema. Each run gets its own pool (`getDatabase` caches one pool per
   identity in-process, which two real processes would not share).
+
+## Index and flag details
+
+- INVALID indexes — the stump a cancelled or timed-out
+  `CREATE INDEX CONCURRENTLY` leaves, which `pg_indexes` still reports as
+  present — are detected via `pg_index.indisvalid` and dropped before the
+  rebuild.
+- `migrations.postgres.useConcurrently: false` vetoes the flag; index DDL then
+  runs inside the atomic transaction (still bounded by the timeouts).
+- Without the flag, a batch containing explicit `CONCURRENTLY` DDL is rejected
+  before the transaction opens — PostgreSQL cannot run it there.

package/dist/{AuditLogCollection-BBbvr8Qc-CoDs0BaX.js → AuditLogCollection-BBbvr8Qc-mcBSLXW9.js}

@@ -1,5 +1,5 @@
 import { t as __exportAll } from "./rolldown-runtime-8H4AJuhK.js";
-import { i as tenantId, r as TenantScoped } from "./dist-BJABhykg.js";
+import { i as tenantId, r as TenantScoped } from "./dist-CivxAJ6C.js";
 import { n as __exportAll$1 } from "./NostrIdentity-RJaATTCt-BbRHWyee.js";
 import { SmrtCollection, SmrtObject, field, foreignKey, smrt } from "@happyvertical/smrt-core";
 //#region ../profiles/dist/chunks/AuditLogCollection-BBbvr8Qc.js

package/dist/{OidcIdentity-COeMidoC-B-cVovG3.js → OidcIdentity-COeMidoC-ETkpZbIS.js}

@@ -42,7 +42,7 @@ var OidcIdentity = class extends SmrtObject {
 	* Find identity by issuer and subject
 	*/
 	static async findBySubject(issuer, subject, options = {}) {
-		const { OidcIdentityCollection } = await import("./OidcIdentityCollection-D4DIZv0K-CexIL7No.js").then((n) => n.r).then((n) => n.r);
+		const { OidcIdentityCollection } = await import("./OidcIdentityCollection-D4DIZv0K-Do3bwRuw.js").then((n) => n.r).then((n) => n.r);
 		return await (await OidcIdentityCollection.create(options)).findBySubject(issuer, subject);
 	}
 	/** Build the collision-free natural key for one OIDC issuer subject. */
@@ -65,7 +65,7 @@ var OidcIdentity = class extends SmrtObject {
 	static async findOrCreate(profile, oidcData, options = {}) {
 		const profileId = profile.id;
 		if (typeof profileId !== "string" || !profileId) throw new Error("OidcIdentity.findOrCreate() requires a saved Profile.");
-		const { reuseExistingOidcIdentityForProfile } = await import("./resolveIdentity-j2WFuKDT-C9XtigVZ.js").then((n) => n.t).then((n) => n.i);
+		const { reuseExistingOidcIdentityForProfile } = await import("./resolveIdentity-j2WFuKDT-BieMIOLz.js").then((n) => n.t).then((n) => n.i);
 		const profileOptions = profile.options ?? {};
 		return (await reuseExistingOidcIdentityForProfile(profileId, {
 			email: oidcData.email,

package/dist/{OidcIdentityCollection-D4DIZv0K-CexIL7No.js → OidcIdentityCollection-D4DIZv0K-Do3bwRuw.js}

@@ -1,6 +1,6 @@
 import { t as __exportAll } from "./rolldown-runtime-8H4AJuhK.js";
 import { n as __exportAll$1 } from "./NostrIdentity-RJaATTCt-BbRHWyee.js";
-import { t as OidcIdentity } from "./OidcIdentity-COeMidoC-B-cVovG3.js";
+import { t as OidcIdentity } from "./OidcIdentity-COeMidoC-ETkpZbIS.js";
 import { SmrtCollection } from "@happyvertical/smrt-core";
 //#region ../profiles/dist/chunks/OidcIdentityCollection-D4DIZv0K.js
 var OidcIdentityCollection_D4DIZv0K_exports = /* @__PURE__ */ __exportAll({