Cc

.env.production.example

@@ -8,7 +8,6 @@ ATELIER_REQUIRE_AUTH=true
 ATELIER_API_KEY=replace-me
 ATELIER_SERVICE_URL=http://atelier:8787
 ATELIER_MCP_MODE=remote
-ATELIER_OPENMEMORY_ENABLED=false
 ATELIER_OPENMEMORY_MCP_SERVER_NAME=openmemory
 ATELIER_LESSON_PR_BOT_ENABLED=false
 GITHUB_TOKEN=

.githooks/pre-push

@@ -18,7 +18,7 @@ echo "📍 Checking Pushing $CURRENT_BRANCH to $REMOTE_BRANCH"
 # Skip session import + external reports — those are slow, code-independent, and
 # covered by tests; only host integrations need to be verified at push time.
 echo "🛠️ Running make install (host integrations only)..."
-if ! ATELIER_NO_IMPORT=1 ATELIER_STRICT=1 make install; then
+if ! ATELIER_STRICT=1 make install; then
     echo -e "\033[1;31m❌ make install failed. Push blocked.\033[0m"
     echo "   Fix the install error locally or use 'git push --no-verify' if you intentionally need to bypass the hook."
     exit 1

.github/copilot-instructions.md

@@ -0,0 +1,46 @@
+---
+applyTo: "**"
+---
+
+<!-- Generated by `scripts/sync_agent_context.py`. Edit `../docs/agent-os/README.md` and rerun the generator. -->
+
+
+## Tool substitution — mandatory
+
+| Use this | Instead of | Why |
+|---|---|---|
+| `mcp__atelier__read` | `Read` / `Bash(cat ...)` | outline-first, 80-95% fewer tokens on large files |
+| `mcp__atelier__search` | `Grep` / `Glob` / `Bash(grep/rg ...)` | ranked, budget-capped, ~280k tokens saved vs naive scan |
+| `mcp__atelier__edit` | `Edit` / `Write` | atomic multi-file, snapshot/rollback, diff recorded |
+| `mcp__atelier__shell` | `Bash(...)` | ANSI-stripped, line-truncated, token-compact |
+| `mcp__atelier__code op=search` | `Bash(grep -rn symbol ...)` | SCIP-indexed, zero subprocess cost |
+
+**Bash is only for git commands and process management.** Do NOT use `Bash(cat file)`, `Bash(grep ...)`, or `Bash(find ...)` — use the atelier equivalents above.
+
+Treat this file as a thin entrypoint. The live source of truth is:
+
+- [Agent OS](../docs/agent-os/README.md)
+- [Workflow](../docs/agent-os/workflow.md)
+- [Taste invariants](../docs/agent-os/taste-invariants.md)
+- [Validation matrix](../docs/agent-os/validation-matrix.md)
+- [Architecture](../docs/architecture/README.md)
+- [Quality scorecard](../docs/quality/scorecard.md)
+- [Plans](../docs/plans/README.md)
+- [Decisions](../docs/decisions/README.md)
+
+## Budget optimizer
+
+- Before changing files, name the deliverable and summarize the smallest viable plan.
+- Keep context narrow: use only the current goal, relevant files, failing command/output, and known constraints.
+- Restate working context in under 10 bullets before editing or after compaction.
+- If more than 10 minutes pass without an edit, name the expected deliverable or check with the user.
+- If the same approach fails twice, call `rescue` or change approach; do not retry a third time.
+
+## Native fallback
+
+If an Atelier MCP tool returns `noop`, is hidden, or is unavailable, use Copilot or VS Code native file reads, workspace search, shell `rg`, or `grep`.
+Always return findings instead of waiting for tool availability to improve.
+
+## Validation
+
+See [Validation matrix](../docs/agent-os/validation-matrix.md) for the minimum checks by change surface. Use the VS Code tasks for repeatable preflight, worktree, and runtime evidence loops.

.gitignore

@@ -58,6 +58,8 @@ coverage.xml
 # Claude Code
 .codex/
 .claude/
+.antigravity/
+.antigravitycli/
 
 # Claude Desktop
 .claude/
@@ -68,9 +70,8 @@ coverage.xml
 # Gemini CLI
 .gemini/
 
-# Github copilot (generated by install, not source)
-.github/copilot-instructions.md*
-.github/chatmodes/
+# Github copilot workspace backups
+.github/copilot-instructions.md.atelier-backup.*
 
 # VSCode test
 .vscode-test/
@@ -141,6 +142,7 @@ results/
 *.sql
 !src/**/*.sql
 *.sqlite3
+vector_cache.sqlite
 dump.rdb
 *.rdb
 

.knowledge/blocks/canonical-identifier-over-display-name.md

@@ -0,0 +1,40 @@
+# Canonical Identifier Over Display Name
+
+- **id:** `canonical-identifier-over-display-name`
+- **domain:** `state.change`
+- **status:** `active`
+- **task_types:** integration_change, data_write, rollback
+
+## Situation
+Human-readable labels such as titles, URLs, paths, and display names can drift. Mutations and rollbacks should target a stable canonical identifier.
+
+
+## Triggers
+- slug
+- handle
+- title
+- url
+
+## Dead ends
+- resolve target from url slug alone
+- use display name as stable identity
+
+## Procedure
+1. Resolve the target through its canonical stable identifier.
+2. Record that identifier before the write.
+3. Use the same identifier for the mutation and the readback.
+
+## Verification
+- A canonical identifier was recorded before the write.
+- The same identifier was used for readback.
+
+## Failure signals
+- wrong target updated
+- ambiguous match set
+
+## When not to apply
+Pure read-only exploration where no state mutation or rollback will happen.
+
+
+## Scope
+- tool_patterns: api.write, db.write

.knowledge/blocks/coding-custom-my-custom-block-5e17f4.md

@@ -0,0 +1,11 @@
+# My Custom Block
+
+- **id:** `coding-custom-my-custom-block-5e17f4`
+- **domain:** `coding.custom`
+- **status:** `active`
+
+## Situation
+Use this when a custom situation arises
+
+## Procedure
+1. Do the thing

.knowledge/blocks/rb-runtime-learned.md

@@ -0,0 +1,19 @@
+# Runtime Learned Recovery
+
+- **id:** `rb-runtime-learned`
+- **domain:** `Agent.shopify.publish`
+- **status:** `active`
+
+## Situation
+When runtime traces show repeated publish failures
+
+## Triggers
+- publish
+- shopify
+
+## Dead ends
+- Retry blindly without validating identifiers
+
+## Procedure
+1. Confirm Product GID
+2. Run validation before retry

.knowledge/blocks/template_alembic-migrations.md

@@ -0,0 +1,38 @@
+---
+id: template-python-fastapi-alembic-migrations
+title: Alembic Migration Discipline
+domain: coding.python-fastapi
+task_types:
+  - feature
+  - refactor
+triggers:
+  - Alembic migration
+  - database schema change
+  - SQLAlchemy model change
+file_patterns:
+  - "**/alembic/**/*.py"
+  - "**/migrations/**/*.py"
+  - "**/models.py"
+tool_patterns:
+  - alembic
+situation: "A code change adds, removes, renames, or changes persistence fields. TODO: replace with your migration review checklist."
+dead_ends:
+  - "Changing ORM models without a matching migration."
+  - "Using autogenerated migration output without reviewing indexes, defaults, and data backfills."
+procedure:
+  - "Create a migration in the same change as the model update."
+  - "Review autogenerated operations for destructive changes and explicit server defaults."
+  - "Keep data backfills idempotent or split them from schema-only migrations when needed."
+  - "Document any manual deployment ordering in the migration or PR description."
+verification:
+  - "Run Alembic upgrade on a clean database."
+  - "Run Alembic downgrade when the project requires reversible migrations."
+  - "Run tests that exercise the changed model fields."
+failure_signals:
+  - "Runtime model expects a column that the migration does not create."
+  - "Migration drops or renames data without an explicit review note."
+required_rubrics: []
+when_not_to_apply: "Read-only query changes that do not alter persisted schema."
+---
+
+TODO: Add your production rollback expectations before using this block unchanged.

.knowledge/blocks/template_dependency-injection.md

@@ -0,0 +1,35 @@
+---
+id: template-python-fastapi-dependency-injection
+title: FastAPI Dependency Injection Boundaries
+domain: coding.python-fastapi
+task_types:
+  - feature
+  - refactor
+triggers:
+  - Depends
+  - request scoped dependency
+  - database session
+file_patterns:
+  - "**/api/**/*.py"
+  - "**/dependencies.py"
+tool_patterns: []
+situation: "A route, service, or test needs access to request-scoped resources such as sessions, auth principals, config, or clients. TODO: document your approved dependency names."
+dead_ends:
+  - "Creating database sessions or network clients directly inside route handlers."
+  - "Using module-level mutable singletons for request-scoped state."
+procedure:
+  - "Expose request-scoped resources through small FastAPI dependency functions."
+  - "Keep dependencies composable: authentication, session, and external clients should be separately overrideable in tests."
+  - "Use dependency_overrides in tests instead of monkeypatching global state."
+  - "Close or yield-clean up resources in the dependency that created them."
+verification:
+  - "Run tests that override the dependency and prove the route uses the override."
+  - "Check that resources created by yield dependencies are cleaned up after requests."
+failure_signals:
+  - "Tests need network access because dependencies cannot be overridden."
+  - "Connection/session lifecycle is hidden in route code."
+required_rubrics: []
+when_not_to_apply: "Simple pure functions that do not depend on request state or external resources."
+---
+
+TODO: Add company-specific dependency override examples here.

.knowledge/blocks/template_pydantic-api-boundaries.md

@@ -0,0 +1,35 @@
+---
+id: template-python-fastapi-pydantic-api-boundaries
+title: Pydantic Models At API Boundaries
+domain: coding.python-fastapi
+task_types:
+  - feature
+  - refactor
+triggers:
+  - FastAPI request model
+  - Pydantic response model
+  - API schema change
+file_patterns:
+  - "**/api/**/*.py"
+  - "**/schemas.py"
+tool_patterns: []
+situation: "A FastAPI endpoint accepts or returns structured data. TODO: adapt examples to your service naming conventions."
+dead_ends:
+  - "Returning raw ORM objects directly from handlers."
+  - "Letting dictionaries drift without a Pydantic schema."
+procedure:
+  - "Define explicit Pydantic request and response models for the endpoint boundary."
+  - "Keep persistence models separate from public API models unless the team has approved coupling."
+  - "Set response_model on the route and verify optional/null fields match the public contract."
+  - "When changing a model, update OpenAPI snapshots or schema tests if the repo has them."
+verification:
+  - "Run the endpoint tests that exercise serialization and validation failures."
+  - "Inspect the generated OpenAPI schema for changed public fields."
+failure_signals:
+  - "Response contains private database fields."
+  - "422 behavior changes without an explicit test."
+required_rubrics: []
+when_not_to_apply: "Pure internal helper changes that do not cross the API boundary."
+---
+
+TODO: Replace this note with examples from your service once the template is adopted.

.knowledge/blocks/template_pytest-layout.md

@@ -0,0 +1,36 @@
+---
+id: template-python-fastapi-pytest-layout
+title: Pytest Layout For FastAPI Services
+domain: coding.python-fastapi
+task_types:
+  - test
+  - feature
+triggers:
+  - FastAPI test client
+  - pytest fixture
+  - endpoint test
+file_patterns:
+  - "tests/**/*.py"
+  - "**/conftest.py"
+tool_patterns:
+  - pytest
+situation: "A change adds or updates tests around FastAPI routes, dependencies, or service behavior. TODO: align fixture names with your repository."
+dead_ends:
+  - "Sharing mutable fixture state across tests without reset."
+  - "Testing route internals instead of public request/response behavior."
+procedure:
+  - "Use a client fixture or factory that applies dependency overrides explicitly."
+  - "Keep database fixtures isolated per test or roll them back predictably."
+  - "Assert status codes and response bodies for both success and validation failure paths."
+  - "Prefer narrow service fixtures over large app-wide fixtures when the endpoint is not under test."
+verification:
+  - "Run the focused test file first, then the broader API test subset."
+  - "Confirm test order does not affect the result."
+failure_signals:
+  - "Tests pass alone but fail when the whole suite runs."
+  - "Fixture setup performs real external calls."
+required_rubrics: []
+when_not_to_apply: "Unit tests for pure functions that do not need the app or test client."
+---
+
+TODO: Link to your preferred test command and fixture naming standard.

.knowledge/blocks/template_request-tracing.md

@@ -0,0 +1,38 @@
+---
+id: template-python-fastapi-request-tracing
+title: Request Tracing And Correlation IDs
+domain: coding.python-fastapi
+task_types:
+  - feature
+  - debug
+  - ops
+triggers:
+  - correlation ID
+  - request ID
+  - tracing
+  - middleware
+file_patterns:
+  - "**/middleware*.py"
+  - "**/api/**/*.py"
+  - "**/*.py"
+tool_patterns: []
+situation: "A request path, middleware, or downstream client needs traceability across service boundaries. TODO: name your canonical header and trace field."
+dead_ends:
+  - "Generating a new request ID when a trusted upstream ID already exists."
+  - "Dropping correlation IDs before downstream calls or background jobs."
+procedure:
+  - "Read the approved inbound correlation header or create one at the service edge."
+  - "Attach the ID to request-local context used by logs and errors."
+  - "Propagate the ID to approved downstream calls and background tasks."
+  - "Return or expose the ID according to the service's debugging policy."
+verification:
+  - "Test requests with and without an inbound correlation ID."
+  - "Inspect logs or response headers to confirm the same ID is preserved."
+failure_signals:
+  - "A single user request produces unrelated IDs across logs."
+  - "Background work cannot be tied back to the originating request."
+required_rubrics: []
+when_not_to_apply: "Offline scripts and one-shot migrations that do not handle service requests."
+---
+
+TODO: Add your trace header name and any privacy constraints.

.knowledge/blocks/template_ruff-black-formatting.md

@@ -0,0 +1,39 @@
+---
+id: template-python-fastapi-ruff-black-formatting
+title: Ruff And Black Formatting Gate
+domain: coding.python-fastapi
+task_types:
+  - feature
+  - refactor
+  - test
+triggers:
+  - Ruff
+  - Black
+  - formatting
+  - lint
+file_patterns:
+  - "**/*.py"
+  - "pyproject.toml"
+tool_patterns:
+  - ruff
+  - black
+situation: "A Python code change is ready for validation. TODO: replace command examples with your repo's make targets."
+dead_ends:
+  - "Declaring completion before local formatting/lint checks run."
+  - "Hand-formatting around a formatter instead of using the configured tools."
+procedure:
+  - "Run the repository's formatter command or Black-compatible target."
+  - "Run Ruff linting on the touched files or project subset."
+  - "Fix reported issues in source rather than suppressing rules casually."
+  - "If a suppression is necessary, keep it narrow and explain it in code review."
+verification:
+  - "Confirm formatter reports no diff."
+  - "Confirm Ruff exits successfully on the touched files or documented subset."
+failure_signals:
+  - "CI formatting check changes files after the agent reports done."
+  - "A broad noqa or per-file ignore hides unrelated issues."
+required_rubrics: []
+when_not_to_apply: "Documentation-only changes with no Python source edits."
+---
+
+TODO: Add your exact lint and format commands.