mainsequence 5.0.1__tar.gz → 5.0.2__tar.gz

{mainsequence-5.0.1/mainsequence.egg-info → mainsequence-5.0.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mainsequence
-Version: 5.0.1
+Version: 5.0.2
 Summary: Main Sequence SDK 
 Author-email: Main Sequence GmbH <dev@main-sequence.io>
 License-Expression: LicenseRef-MainSequence-Proprietary

{mainsequence-5.0.1 → mainsequence-5.0.2}/agent_scaffold/AGENTS.md

@@ -40,8 +40,8 @@ Compare the SDK value with the installed version reported by
 Refresh the managed scaffold only when `PINNED_FROM.txt` is missing or its
 `pinned_version` differs from the installed SDK version:
 
-- `mainsequence project update AGENTS.md --path .`
 - `mainsequence project update_agent_skills --path .`
+- `mainsequence project update AGENTS.md --path .`
 
 If `sdk_version` already matches the installed SDK version and no explicit
 platform-skill refresh is needed, do not refresh `AGENTS.md` or
@@ -76,8 +76,7 @@ Core responsibilities:
 - translate user business logic into reusable code under `src/` so it can be reused by APIs,
   dashboards, jobs, and other project components instead of duplicating logic in integration
   layers
-- maintain the repository through explicit `.agents/` project-state files and the bug auditor skill
-  for blocker and SDK/platform issue assessment
+- use the bug auditor skill for blocker and SDK/platform issue assessment
 
 Typical outcomes include:
 
@@ -89,7 +88,7 @@ Typical outcomes include:
   Center surface before building or deploying it
 - build reusable business logic in `src/` and keep thin integration layers in APIs, jobs, and
   dashboards
-- keep the repository auditable through project-state records, blocker tracking, and bug reporting
+- assess blockers and suspected SDK or platform issues from concrete repository and runtime evidence
 
 Working rules for this role:
 
@@ -117,13 +116,6 @@ User-resolution rule for agents:
 
 Delegation rules:
 
-- when work is delegated or queued for later, write the task in `.agents/tasks.md` according to
-  the skill that should execute it
-- each delegated or queued task in `.agents/tasks.md` must state:
-  - the exact task scope
-  - the owning skill
-  - the expected output, decision, or artifact
-  - the validation or evidence required before it can be marked complete
 - if delegation is available, declare for each sub-agent:
   - the exact task it owns
   - the exact skill or skills it must use
@@ -190,6 +182,9 @@ Typical routing:
 - project setup, local checkout, CLI environment, scaffolding, and standard
   repository layout:
   `.agents/skills/mainsequence/sdk_project_execution/SKILL.md`
+- local environment repair, project authentication refresh, SDK updates,
+  managed skill refresh, and canonical project sync:
+  `.agents/skills/mainsequence/maintenance/project-maintenance/SKILL.md`
 - turning an existing project into a project-backed coding agent, defining
   project-owned skills, and authoring `.agents/agent_card.json`:
   `.agents/skills/mainsequence/project_to_agent/SKILL.md`
@@ -234,31 +229,28 @@ For any non-trivial Main Sequence task:
 
 1. Read the latest relevant Main Sequence documentation.
 2. Compare the implementation against the latest documented behavior.
-
-4. Check `.agents/tasks.md` for current priorities.
-
-7. Confirm you are in the correct project checkout, or use `--path` explicitly.
-8. Confirm platform context with:
+3. Confirm you are in the correct project checkout, or use `--path` explicitly.
+4. Confirm platform context with:
    `mainsequence project current --debug`
-9. Before validations or live checks, run:
+5. Before validations or live checks, run:
    `mainsequence project refresh_token --path .`
-10. If git push or pull is required, use:
+6. If git push or pull is required, use:
     `mainsequence project open-signed-terminal <PROJECT_ID>`
-11. Before proceeding with non-trivial Main Sequence work, update the project SDK:
+7. Before proceeding with non-trivial Main Sequence work, update the project SDK:
     `mainsequence project update-sdk --path .`
-12. After updating the SDK, compare `mainsequence --version` with
+8. After updating the SDK, compare `mainsequence --version` with
     `.agents/skills/mainsequence/PINNED_FROM.txt` field `sdk_version=...`
     (`pinned_version=...` is its compatibility alias), and verify that the
     sentinel contains platform manifest and resource hashes.
-13. If `PINNED_FROM.txt` is missing, `sdk_version` differs from the installed
+9. If `PINNED_FROM.txt` is missing, `sdk_version` differs from the installed
     SDK version, or a platform-skill refresh is explicitly required, refresh
     the managed scaffold files:
-    `mainsequence project update AGENTS.md --path .`
     `mainsequence project update_agent_skills --path .`
-14. If `sdk_version` already matches the installed SDK version and no platform
+    `mainsequence project update AGENTS.md --path .`
+10. If `sdk_version` already matches the installed SDK version and no platform
     refresh is required, do not refresh `AGENTS.md` or
     `.agents/skills/mainsequence/` as a startup ritual.
-15. Verify platform state with the CLI or platform tooling instead of guessing.
+11. Verify platform state with the CLI or platform tooling instead of guessing.
 
 ## Orchestrator Rule
 
@@ -276,15 +268,13 @@ project behavior exists and has been verified. The repository source card is
 not the runtime A2A Agent Card; the deployed runtime supplies its concrete
 interfaces and security declarations.
 
-Before the final response:
-
-- update the relevant `.agents/` project-state files whenever project understanding, verified
-  state, open tasks, blockers, or historical record changed during the turn
-
 Use `.agents/skills/mainsequence/project_design/SKILL.md` as the platform
 source of truth for intent and ontology. Use
 `.agents/skills/mainsequence/sdk_project_execution/SKILL.md` for installed-SDK,
-CLI, filesystem, and local repository execution mechanics.
+CLI, filesystem, and local repository execution mechanics. Use
+`.agents/skills/mainsequence/maintenance/project-maintenance/SKILL.md` for
+repeatable environment, authentication, SDK, scaffold-refresh, and project-sync
+routines.
 
 ## Core Working Rules
 
@@ -295,8 +285,8 @@ CLI, filesystem, and local repository execution mechanics.
 - do not hide failures
 - record the exact failing step, command, or workflow
 - when hitting a roadblock, blocker, or error, report it back to the user clearly and promptly
-- if local code or local docs conflict with the latest Main Sequence docs, record the discrepancy
-  and create follow-up work
+- if local code or local docs conflict with the latest Main Sequence docs, report the discrepancy
+  and the concrete next action
 - when unsure, verify
 - if the active virtual environment is missing libraries that are already declared in
   `requirements.txt`, install those missing libraries into the virtual environment before
@@ -359,7 +349,7 @@ Rules:
 
 If something may be a Main Sequence SDK, documentation, or platform issue:
 
-- record what failed
+- report what failed
 - explain why it may be a Main Sequence issue
 - suggest a concrete improvement
 
@@ -370,10 +360,4 @@ If something may be a Main Sequence SDK, documentation, or platform issue:
 - prefer explicit facts over vague statements
 - surface failures early
 - distinguish verified facts from assumptions
-
-## Project-State Files Under `.agents/`
-
-
-Do not improvise their meaning in domain skills. Use the project bootstrap skill rules to reconcile
-them after material work.
 <!-- mainsequence-agent-scaffold:end -->

{mainsequence-5.0.1 → mainsequence-5.0.2}/agent_scaffold/skills/maintenance/bug_auditor/SKILL.md

@@ -20,8 +20,7 @@ This skill is for diagnosis and assessment. Default behavior is read-only unless
 ## This Skill Can Do
 
 - inspect project state and summarize completion status
-- read `.agents/tasks.md` and `.agents/status.md` first when they exist
-- use repo state, logs, test output, stderr, and task files as evidence
+- use repo state, logs, test output, and stderr as evidence
 - classify failures into:
   - target-project issue
   - environment or credentials issue
@@ -62,6 +61,9 @@ This skill audits. It does not implement by default.
   `.agents/skills/mainsequence/platform_operations/access_control_and_sharing/SKILL.md`
 - Streamlit dashboard deployment or release issues:
   `.agents/skills/mainsequence/platform_operations/orchestration_and_releases/SKILL.md`
+- environment repair, authentication refresh, SDK updates, managed skill
+  refresh, and project sync after the failure is classified:
+  `.agents/skills/mainsequence/maintenance/project-maintenance/SKILL.md`
 
 Streamlit dashboard design and implementation failures are target-project application code issues unless the evidence points to Main Sequence deployment, resource discovery, image selection, release creation, or platform runtime behavior.
 
@@ -69,11 +71,7 @@ Streamlit dashboard design and implementation failures are target-project applic
 
 1. `AGENTS.md`
 2. `.agents/skills/mainsequence/project_design/SKILL.md`
-3. `.agents/tasks.md` when it exists
-4. `.agents/status.md` when it exists
-5. `.agents/record.md` when stable references or project ids matter
-6. `.agents/journal.md` when repeated failures or prior investigations may be relevant
-7. the latest relevant Main Sequence docs for the failing workflow
+3. the latest relevant Main Sequence docs for the failing workflow
 
 ## Inputs This Skill Needs
 
@@ -101,17 +99,15 @@ For every audit, decide:
 
 ### 1. Stay read-only unless edits were explicitly requested
 
-Do not modify code, docs, or project-state files unless the task explicitly asks for edits.
+Do not modify code or docs unless the task explicitly asks for edits.
 
-### 2. Start with project-state files
+### 2. Start with current evidence
 
-Read `.agents/tasks.md` and `.agents/status.md` first when they exist.
-
-Use them as hypotheses, not as proof.
+Inspect the repository state, available logs, failing commands, stderr, and test output.
 
 ### 3. Use evidence, not impressions
 
-Use repo state, logs, test output, stderr, and task files as evidence.
+Use repo state, logs, test output, and stderr as evidence.
 
 When reporting a blocker or failure, include:
 
@@ -127,7 +123,6 @@ Before each major investigation step, emit a short progress update that says wha
 
 Especially announce when you are:
 
-- reading `.agents/tasks.md` or `.agents/status.md`
 - inspecting a failing command, traceback, or stderr excerpt
 - checking the local `mainsequence` package or version
 - inspecting or cloning the public `mainsequence-sdk` repository
@@ -184,7 +179,6 @@ When reviewing an audit, look for:
 
 Do not claim audit completion until you have checked:
 
-- `.agents/tasks.md` and `.agents/status.md` first when they exist
 - the overall state is one of:
   - `finished`
   - `in_progress`

mainsequence-5.0.2/agent_scaffold/skills/maintenance/project-maintenance/SKILL.md

@@ -0,0 +1,200 @@
+---
+name: project-maintenance
+description: Maintain an existing Main Sequence project checkout using the SDK-version-matched CLI. Use for inspecting project state, building or repairing .venv, refreshing local project authentication, updating the SDK, refreshing SDK and platform skills plus AGENTS.md, publishing changes with project sync, or diagnosing a partially completed maintenance workflow.
+---
+
+# Main Sequence Project Maintenance
+
+Maintain the local project through the installed Main Sequence CLI. Treat the
+CLI as the canonical implementation of SDK-version-specific filesystem, `uv`,
+authentication, scaffold, and Git behavior. Select and sequence commands here;
+do not reproduce their Python implementation or replace them with ad hoc shell
+workflows.
+
+## Preserve The Boundary
+
+Own:
+
+- local project inspection and maintenance sequencing;
+- `.venv`, SDK, managed skill, `AGENTS.md`, and Git-release workflows;
+- precondition checks, explicit approval gates, and partial-failure diagnosis.
+
+Do not own:
+
+- platform ontology or Project Blueprint design;
+- domain implementation for MetaTables, DataNodes, APIs, jobs, or releases;
+- backend repository reconciliation;
+- MCP authentication or access-token extraction;
+- project-owned skills outside `.agents/skills/mainsequence/`.
+
+Route architecture changes to `project_design`, implementation routing to
+`sdk_project_execution`, and failure classification to
+`maintenance/bug_auditor` when diagnosis extends beyond the maintenance
+workflow itself.
+
+## Inspect Before Changing State
+
+1. Read `AGENTS.md` and the relevant repository skills.
+2. Confirm the repository root and the user's requested maintenance outcome.
+3. Inspect project and Git state:
+
+   ```bash
+   mainsequence project current --debug --json
+   mainsequence project sdk-status --path . --json
+   git status --short
+   ```
+
+4. Separate existing user changes from changes created by the maintenance
+   task. Never assume every uncommitted file belongs to the current task.
+5. Read command help when the installed SDK differs from the instructions in
+   this skill. The installed CLI is authoritative for its version.
+
+If `mainsequence` is unavailable, stop and report that the SDK CLI must be
+installed or provided by the local development integration. Do not use an MCP
+bearer token or an unpinned downloaded CLI as a substitute.
+
+## Build Or Repair The Local Environment
+
+Build the project environment with:
+
+```bash
+mainsequence project build_local_venv --path .
+```
+
+The command reads the project Python requirement, resolves `uv`, creates
+`.venv`, and synchronizes dependencies. Do not manually parse
+`pyproject.toml` or reconstruct those steps.
+
+When an existing `.venv` is incompatible, show the detected mismatch and ask
+before replacing it. Only then run:
+
+```bash
+mainsequence project build_local_venv --path . --recreate
+```
+
+Treat `.venv` as generated state, never as source code or durable project
+documentation.
+
+## Refresh Project Authentication
+
+Refresh only the CLI-managed runtime authentication entries in the project
+`.env`:
+
+```bash
+mainsequence project refresh_token --path .
+```
+
+Require normal Main Sequence CLI login first. Never print, inspect, summarize,
+copy, or return access and refresh token values. Do not attempt to extract the
+calling MCP host's protected bearer token. If authentication is unavailable,
+stop and request login through the normal CLI or development integration.
+
+## Update The Project SDK
+
+Inspect the current status, preview the update, and then update when requested
+or required by repository instructions:
+
+```bash
+mainsequence project sdk-status --path . --json
+mainsequence project update-sdk --path . --dry-run
+mainsequence project update-sdk --path .
+```
+
+`update-sdk` updates the lock and local environment. It does not publish the
+working tree. After the SDK changes, refresh the managed agent context and run
+the repository-specific validation required by `AGENTS.md`.
+
+Do not automatically commit or push an SDK update unless the user also asked
+to publish the project changes.
+
+## Refresh Managed Skills And Instructions
+
+Refresh the SDK-owned and platform-owned skills first, then update the managed
+Main Sequence block in `AGENTS.md`:
+
+```bash
+mainsequence project update_agent_skills --path .
+mainsequence project update AGENTS.md --path .
+```
+
+Verify `.agents/skills/mainsequence/PINNED_FROM.txt` after success:
+
+- `sdk_version` must match the installed project SDK;
+- platform manifest, ontology, and resource hashes must be present;
+- project-owned skills outside `.agents/skills/mainsequence/` must remain
+  untouched.
+
+The skill update is staged and atomic. If it fails, report the failing lane and
+preserve the previous valid managed tree. Because this operation can update
+this skill, reload the refreshed `project-maintenance/SKILL.md` before starting
+another maintenance routine.
+
+## Publish With Canonical Project Sync
+
+Use project sync only when the user intends to commit, tag, and push all
+reviewed project changes.
+
+Before execution:
+
+```bash
+git status --short
+git diff --stat
+mainsequence project sync --path . -m "<specific commit message>" --dry-run
+```
+
+Review every pending file because the command stages with `git add -A`. Do not
+continue when unrelated or unexplained changes would be included.
+
+After review:
+
+```bash
+mainsequence project sync --path . -m "<specific commit message>"
+```
+
+The canonical command always:
+
+1. applies a patch version bump;
+2. runs `uv lock`;
+3. runs `uv sync`;
+4. exports locked production requirements;
+5. stages and commits the changes;
+6. creates an annotated `v<version>` tag; and
+7. pushes the branch and tag with `--follow-tags`.
+
+Do not offer alternate bump modes, a no-push mode, or a hand-written sequence
+of equivalent commands. Do not call `sync-after-commit` or any backend repair
+endpoint. The GitHub branch-push webhook owns backend repository
+reconciliation.
+
+## Diagnose Partial Completion Before Retrying
+
+Never rerun an entire mutating workflow blindly.
+
+- After an environment failure, inspect `.venv`, the declared Python
+  requirement, and the exact failing `uv` output.
+- After an SDK update failure, inspect `pyproject.toml`, `uv.lock`, the active
+  environment, and the failed command before retrying.
+- After a skill refresh failure, keep the previous managed tree and identify
+  whether the SDK or platform lane failed.
+- After a sync failure, inspect `git status`, `git log -1`, the current project
+  version, tags pointing at `HEAD`, the upstream branch, and remote tags before
+  deciding which step remains.
+
+A blind sync retry may create another patch bump or collide with an existing
+commit or tag.
+
+## Validate And Report
+
+Run only repository-relevant validation discovered from `AGENTS.md`, project
+documentation, and the changed files. Report:
+
+- the maintenance routine performed;
+- the project path and installed SDK version;
+- files or generated state changed;
+- validation run and its result;
+- whether changes remain local or were committed, tagged, and pushed;
+- any remaining authentication, environment, Git, or webhook verification
+  gap.
+
+Never claim backend reconciliation succeeded solely because the local push
+succeeded. Verify backend state separately when the task requires that claim.

{mainsequence-5.0.1 → mainsequence-5.0.2}/agent_scaffold/skills/sdk_project_execution/SKILL.md

@@ -65,6 +65,9 @@ Do not let this skill become a domain manual.
   `.agents/skills/mainsequence/command_center/widgets/app_components/SKILL.md`
 - project audits, blocker analysis, and upstream SDK assessment:
   `.agents/skills/mainsequence/maintenance/bug_auditor/SKILL.md`
+- local environment repair, project authentication refresh, SDK updates,
+  managed skill refresh, and canonical project sync:
+  `.agents/skills/mainsequence/maintenance/project-maintenance/SKILL.md`
 - jobs, schedules, artifacts, images, resources, releases, and Streamlit dashboard deployment:
   `.agents/skills/mainsequence/platform_operations/orchestration_and_releases/SKILL.md`
 - RBAC and sharing:

{mainsequence-5.0.1 → mainsequence-5.0.2}/mainsequence/cli/api.py

@@ -2062,62 +2062,6 @@ def get_project_data_node_updates(
                 os.environ[k] = v
 
 
-def sync_project_after_commit(
-    project_id: int | str, *, timeout: int | None = None
-) -> dict[str, Any] | None:
-    """
-    Notify the backend that a project commit has been pushed.
-
-    Single source of truth:
-      - delegates request behavior and payload parsing to `Project.sync_project_after_commit()`
-    """
-    try:
-        project_uid = resolve_project_uid(project_id)
-        payload = _run_sdk_model_operation(
-            module_name="mainsequence.client.models_foundry",
-            class_name="Project",
-            operation=lambda ClientProject: ClientProject.sync_project_after_commit(
-                project_uid,
-                timeout=timeout,
-            ),
-            project_id_env=resolve_project_row_id(project_id),
-            project_uid_env=project_uid,
-        )
-        if payload is None:
-            return None
-        if isinstance(payload, dict):
-            return payload
-        if hasattr(payload, "model_dump"):
-            return payload.model_dump()
-        return {"id": getattr(payload, "id", None)}
-    except Exception as e:
-        err_name = type(e).__name__
-        if err_name in {"AuthenticationError", "PermissionDeniedError"}:
-            raise NotLoggedIn(str(e) or "Not logged in.") from e
-        if err_name == "NotFoundError":
-            raise ApiError(f"Project not found: {project_id}") from e
-        raise ApiError(f"Project post-commit sync failed: {e}") from e
-
-
-def prime_sync_project_after_commit_sdk() -> None:
-    """
-    Load the SDK modules needed by `sync_project_after_commit()` before the active
-    virtual environment is mutated by `uv sync`.
-
-    `project sync` runs inside the same `.venv` that it later updates. If the
-    environment changes during the command, a fresh `mainsequence.client` import
-    can fail late in the flow. Priming the relevant SDK modules first keeps the
-    post-commit sync path available in the current process.
-    """
-    try:
-        importlib.import_module("mainsequence.client")
-        importlib.import_module("mainsequence.client.utils")
-        importlib.import_module("mainsequence.client.base")
-        importlib.import_module("mainsequence.client.models_foundry")
-    except Exception as e:
-        raise ApiError(f"Project post-commit SDK import failed: {e}") from e
-
-
 def create_project_image(
     *,
     project_repo_hash: str,