softwareone-platform

Only triage and respond to unresolved PR review comments: decide fix versus answer per comment, apply scoped changes, and reply in the thread. Does not create the PR, commit, or push the branch.

Apply one approved dashboard-failure Jira decision: create, update, reopen, merge, or skip an MPT bug with dashboard evidence fields.

Triage MPT extension dashboard failures from App Insights or CSV: prepare a batch review of findings with proposed Jira actions, then apply approved decisions via the dashboard Jira task.

Only create the work branch when starting work: derive the branch name from the Jira issue and apply the repo naming pattern. Does not change Jira state.

When a task needs a low-level Git branch operation for an already-decided branch name — create or switch safely on the correct base. Used by branch tasks, not for deriving names from Jira.

Run the full start-of-work flow for a Jira issue, leaving a work branch created and the issue in active development. Orchestrates the branch and Jira-start tasks; use for feature, bugfix, hotfix, or backport.

Only move Jira state when starting work: transition the issue and its parent chain to In Progress, ensure active-sprint placement, and check reassignment. Does not create a branch.

Only open or update a repository-compliant pull request to publish committed work, reusing the existing PR instead of duplicating. Does not update docs, run checks, or transition Jira.

Run canon-generate's evidence-gathering and drafting concurrently across several objects at once, via the Workflow tool, with one consolidated human checkpoint instead of a per-object one. Never writes directly into objects/ — same draft/promotion boundary as canon-generate.

Generate or refresh a canon object document from real evidence — OpenAPI schema, live multi-Actor API samples (STAGING/PROD), and platform source code — instead of the fully manual authoring session. Always produces a draft for human review; never writes directly into objects/.

Read-only staleness scan — rank canonised objects by how far their source/spec has drifted since a recorded baseline, and print a ranked report. Never drafts, never writes objects/, needs no live tokens. Feeds /canon-drift-update.

Reconcile drift for a canonised object — detect what changed in source/spec since its baseline, then draft a SCOPED edit of the existing canon (touched sections only) for PM review. Draft-only; hands to /canon-submit-pr. Run /canon-drift-report first to see what drifted.

Promote a reviewed canon-generate draft into objects/ and open a PR — branch, commit (draft + any pending bookkeeping edits), push, and gh pr create. Always confirms with the human before any git/GitHub action.

Write and review high-quality Python tests using pytest. Use whenever creating new tests, refactoring existing tests, debugging test failures, improving coverage, mocking dependencies, mocking HTTP calls, or generating test data — even when the user says "tests" without further detail. Enforces pytest-mock over unittest.mock, mandatory type annotations and docstrings, fixtures kept out of test files (in conftest.py or modules imported by it), factories for test data, pytest-httpx for HTTP mocks, ≥90% coverage (target ≥95%), and testing only public interfaces — never private functions or methods.

Use when migrating or syncing code from the private mpt-library-framework into this OSS mpt-framework-net repo — extracting a new component (delta, operation, messagehub, etc.) or propagating later upstream changes. Covers package layout (Abstractions / main / EFCore split), naming (singular nouns, EFCore suffix, Apache-2.0), standard dependency substitutions, what to skip from upstream (Models.Shared, business taxonomy, platform-entity-coupled code), and verification.