task: A-entry — canonical Apps entrypoints and application composition #408

Current delivery decision (2026-11-15)

Use canonical Apps entrypoints and retire old Python module paths in the same PR that moves their implementations. The final tracked package root contains only __init__.py and __main__.py; a build-generated _version.py is separate metadata. Preserve the normal inferrsi command and python -m inferrsi, not indefinite inferrsi.cli / inferrsi.arguments import aliases. This closes root-level scattering, old entrypoints and core dependency boundaries; it does not include a further internal refactor of existing subpackages.

Related architecture: #387 and #371. Shared contracts/resources: #451. Existing-path retirement/core closure: #412. MindIE-specific relocation is separate: #452.

PR checklist

• A1 — Switch to canonical Apps CLI and delete the root CLI modules. Move implementation to the apps/cli/ package (apps/cli/__init__.py exposes the canonical main, apps/cli/arguments.py holds the parser); point the console script to inferrsi.apps.cli:main and make root __main__.py delegate only to that canonical entry. Update active plugins, DSH/Codex/WorkSwarm entrypoints, docs and current tests in this same PR; delete root cli.py and arguments.py. Do not add permanent forwarders or module-registry tricks. Remove the cli.py entry from APPLICATION_ENTRIES in tests/_gate_policy_support.py in this same PR (apps/cli/ is already classified as apps by its first path segment), so a reintroduced root cli.py is not admitted as Apps. Cover the PR Connect report Dashboard and trace to fixed attempt records #450 fixed surfaces in the migrated CLI: status/trace/report/dashboard --record-ref success paths and their error branches (missing --storage-config, fixed report/dashboard without --output, --live with --output, --port without --live), byte-comparing behavior with Connect report Dashboard and trace to fixed attempt records #450's head.
• A2 — Move chart rendering and benchmark application composition. Move summary_chart to apps/views/summary_chart.py and benchmark_operations to apps/benchmark_operations.py. Update Dashboard/reporting/tool/CLI consumers and tests in this same PR; delete both old root files. Do not turn rendering into authoritative evaluation. Run tools/campaign_summary_chart.py as a real entry point in acceptance, remove benchmark_operations.py from APPLICATION_ENTRIES in tests/_gate_policy_support.py so the tuple ends empty, and update the gate tests without relaxing the rule.

Dependencies and incoming work

Preferred queue: C1, C2 (#451), then R1 (#412), then A1 and A2. R1 does not technically require C1/C2 or A1: it can update the current CLI consumers while retiring the existing paths. Keep those as preferred integration order, not artificial architecture blockers. A1 is scheduled after R1 to move an already-clean CLI once; A2 is based on A1's canonical Apps entrypoint. No implementation session starts merely because its read-only plan is complete.

PR #449 is merged. Retain its fixed-record status route and --record-ref, --storage-config, --assessment-ref validation. PR #450 currently adds fixed report/Dashboard/trace surfaces: if it merges first, migrate that actual implementation; if this migration lands first, #450 must target the canonical Apps entrypoints. Do not recreate or overwrite the ContextProvider/candidate-store work in #446/#448. Refresh this observation at implementation time rather than treating open PR code as main.

Acceptance

• Preserve normal command behavior, exit codes and stdout/stderr bytes, including failure cases and the existing host matrix; retire old Python imports explicitly.
• Verify console / python-module / plugin entrypoints and installed-wheel execution outside the checkout, including resources.
• Keep the existing OBS reader/layout decision and evaluator/factual-record authority unchanged; no new evaluate command, warning layer or format retirement.
• Update testing injection points to the canonical implementation while retaining real execution/inspection injection semantics and old-command behavioral comparisons; do not delete coverage to hide regressions.
• Record exact root-file removals and a shrinking tracked-root inventory. Overall isolation_complete remains false until the complete closure in task: R01 — retire legacy root paths and close core dependencies #412 is actually proven.

Delivery discipline

• One implementation Claude session handles only the current PR in an isolated worktree. A second Claude session may only plan the next PR in a separate planning worktree; it must not edit source.
• Move implementation, update all active consumers/docs/tests, and delete the corresponding old root modules in the same PR. Do not introduce permanent compatibility shells or module-registry aliases.
• Preserve CLI command behavior, original evidence bytes, record/format contracts and isolation/security checks. Announce retired Python import/serialization module paths; historical frozen evidence remains pinned to its original commit or wheel, not rewritten.
• Before each implementation or review handoff, refresh main and confirmed dependencies. A completed plan is not implementation or approval. No automatic merge.

Historical context retained

The original scope, approved on 2026-11-12, depended on K01, #397 and OBS, preferably after B01, and required the three-host/installed-wheel matrix. Those prerequisites are now merged. The earlier permanent-forwarder decision for root CLI modules and the old “do not delete shells” restriction are superseded by the revised owner decision above; the command/format/fact/security acceptance requirements are not. #397 remains the standardized Apps output slice, not a vehicle for new migration scope. Existing issue discussion and review history are not rewritten by this checklist update.