Repository navigation
Conversation
…ct-rest-api-implementation # Conflicts: # integrations/dsh/plugins/powercontext/lib/index.js # integrations/dsh/plugins/powercontext/openapi/powercontext.yaml # integrations/dsh/plugins/powercontext/scripts/openapi-ops.mjs # integrations/dsh/plugins/powercontext/src/client.ts # integrations/dsh/plugins/powercontext/src/operations.generated.ts # integrations/dsh/plugins/powercontext/tests/client.spec.ts # integrations/dsh/plugins/powercontext/tests/operations-coverage.spec.ts # integrations/opencode/plugins/powercontext/lib/index.js # integrations/opencode/plugins/powercontext/src/client.ts # integrations/opencode/plugins/powercontext/src/operations.generated.ts # integrations/pi/plugins/powercontext/src/client.ts # integrations/pi/plugins/powercontext/src/operations.generated.ts # openapi/powercontext.yaml # scripts/generate_api.py # scripts/generate_js_operations.py # src/powercontext/builtin/runtime/application.py # src/powercontext/builtin/runtime/relational.py # src/powercontext/client/client.py # src/powercontext/http/__init__.py # src/powercontext/http/_generated/operations.py # src/powercontext/server/app.py # tests/e2e/test_builtin_runtime.py # tests/test_api_contract.py # tests/test_client.py # tests/test_js_operations.py
…ct-rest-api-docs # Conflicts: # zensical.toml
* fix: env file parser Signed-off-by: thunguo <tew@apache.org> * perf(cli): scan multiline env values incrementally * test(cli): decouple multiline env fixture --------- Signed-off-by: thunguo <tew@apache.org>
…ct-rest-api-implementation
…tions/actions/configure-pages-6.0.0 build(deps): bump actions/configure-pages from 5.0.0 to 6.0.0
…tions/apache/skywalking-eyes/header-0.9.0 build(deps): bump apache/skywalking-eyes/header from 0.8.0 to 0.9.0
* feat(pi): add external skill tools * fix(pi): defer full profile promotion
* feat(pi): add artifact candidate review tools * fix(pi): align candidate review schemas * fix(pi): close candidate review schemas * fix(pi): expose revision schema as object
…#1622) * feat(integrations): add layered PowerContext Skill routing * test(openclaw): build runtime before native Skill discovery * test(integrations): reject nested Handoff carriers in evaluation * fix(integrations): unify Skill entries and validate domain reads * fix(integrations): align Handoff workflows and preview boundaries * fix(hermes): expose native Skill discovery metadata * docs(pi): align guidance with candidate review capabilities
* feat(pi): add experience and skill generation tools * fix(pi): preserve generation schema and handle timeouts
* feat(pi): add memory changes and stats tools * fix(pi): preserve nullable memory change cursors
* docs(rfc): design topic memory generic API completion * docs(rfc): add English topic memory API design * docs(rfc): clarify topic memory tag schema migration * docs(rfc): define empty lexical projection behavior
…Codex (oceanbase#1580) * fix(integrations): honor approved transport, status classes, and hook budgets fix(opencode): pass the approved insecure-HTTP consent into the TUI client fix(opencode): stop status polling and in-flight requests when the view unmounts fix(opencode): keep auth, version, unavailable, and invalid outcomes distinct in the statusline fix(opencode): reject a statistics response without required totals instead of showing no data fix(claude-code): reject a statistics response without required totals instead of showing no data fix(codex): keep the Stop hook HTTP budget below the 3-second host deadline fix(integrations): harden scope binding and status diagnostics fix(claude-code): resolve the effective Server URL before loading persisted authorization fix(claude-code): classify statusline failures into typed diagnostics fix(claude-code): bound Scope response reads and git identity by the absolute HTTP budget fix(codex): reject malformed stats responses instead of reporting no data fix(codex): bound Scope response reads by the absolute HTTP budget fix(opencode): omit empty session and workspace binding keys on non-session routes fix(integrations): report token savings through the stats contract fix(codex): POST /v1/stats with a Scope selection instead of the unsupported GET query fix(codex): classify Stop-hook failures into typed diagnostics with cross-invocation dedup fix(codex): classify Scope binding transport, auth, and HTTP status failures fix(claude-code): send the same stats contract and bound both windows by one absolute HTTP budget fix(opencode): resolve TUI statusline and /pc Scope through server-side Scope bindings fix(opencode): accept JSONC TUI config and validate it before replacing any installed file feat(codex): report token savings after turns feat(claude-code): show token savings statusline fix(opencode): color token savings by outcome feat(opencode): show daily and 30d token savings fix(opencode): rename statusline metric to compression proxy chore(opencode): sync tui type declarations fix(opencode): clarify statusline wording fix(opencode): bound statusline refresh with timeouts feat(opencode): expand PowerContext statusline fix(opencode): load token statusline in TUI feat(opencode): show token savings in statusline fix(opencode): register TUI plugin during setup feat(opencode): add PowerContext TUI command * fix(integrations): harden statusline budgets, scope aborts, and install hygiene fix(opencode): stop accumulating stale tui.json entries, preserve JSONC comments, tolerate malformed entries, and drop the duplicate Skill install fix(claude-code): canonicalize Server URLs before matching persisted authorization, tighten statusline ownership, honor per-request Scope timeouts, and bound request opens by wall clock fix(codex): honor per-request Scope timeouts and bound request opens by wall clock fix(opencode-tui): abort in-flight Scope resolution on unmount, keep invalid responses distinct from offline, trim resolved Scope IDs, and avoid cleaning the shared lib output Align compact token formatting between Python and JavaScript, refresh generated lib bundles, and extend regression coverage for budgets, classification, and install hygiene. * fix(integrations): refresh the Claude plugin cache and bound chunked reads fix(claude-code): bump the plugin and marketplace manifests to 0.1.2, refresh an existing marketplace and plugin before configuring the status line, and bind every response read to the absolute deadline fix(codex): bind every response read to the absolute deadline Add slow chunked-extension drip regression tests for the Claude status line and the Codex Stop hook, update the version contract and setup refresh coverage, and align the Claude Code integration docs. * fix(integrations): address PR 1580 review follow-ups * test(opencode): cover TUI file URI parsing
* feat(topic-memory): complete generic Artifact APIs * fix(topic-memory): allow empty analyzer output * fix(topic-memory): attribute indexing usage by scope * fix(topic-memory): isolate usage recording failures * fix(persistence): make legacy tag upgrades retryable
* fix(service): decode Windows native command output with the OEM code page icacls, whoami, and schtasks output was captured with text=True and no explicit encoding. In a no-window process under Python UTF-8 mode the console reports the OEM code page, strict UTF-8 decoding fails inside the subprocess reader thread, and the caller silently receives returncode == 0 with stdout set to None. Protected environment-file validation then crashed with an AttributeError instead of raising a domain error, and ACL checks could be skipped. Route all three commands through run_windows_command, which forces CREATE_NO_WINDOW so the child always uses the system OEM code page, captures bytes, and decodes on the calling thread so decoding failures surface as SubprocessError and are reported as ServiceError or ProtectedEnvironmentFileError. Fixes oceanbase#1627 * fix(service): address review findings on Windows command decoding run_windows_command now treats LookupError as a command failure, so a call from a non-Windows path reports the missing OEM codec through SubprocessError instead of escaping the caller except clause. test_windows_support_preserves_native_command_error mocked every subprocess.run call, so _user_identity failed at whoami and support() never reached the scheduler query; the assertion passed for the wrong reason. The mock now dispatches on the executable and the test asserts the Task Scheduler wording, so it exercises the schtasks decoding path it names. test_service_install_cli_expands_the_environment_file_home_directory set HOME, which Windows expanduser ignores in favour of USERPROFILE. Setting both makes the test pass on Windows; it failed on ee5c076 as well. Refs oceanbase#1627 * ci(windows): run the service command encoding regressions on Windows The six encoding cases added alongside run_windows_command are guarded by skipif(os.name != "nt"). The full suite runs on Linux, which collected and skipped them, and neither Windows job listed these files, so no job exercised the regressions the previous two commits added. Add tests/test_service.py and tests/test_service_environment.py to the windows-unit-portability job, the Windows Python unit test job. Verified locally on Windows that all six cases execute rather than skip. Refs oceanbase#1627 * test(service): own the Windows env-file fixtures before locking their ACL The windows-unit-portability job failed 13 cases in tests/test_service.py and tests/test_service_environment.py: load_protected_environment_file rejected every fixture with "--env-file must be owned by the current user". Hosted Windows runners execute jobs under an elevated token whose default owner is the Administrators group, so a file pytest creates is owned by that group rather than by the account itself, and the loader's owner check refuses it. The failures masked the ACL and encoding assertions that follow. Set the owner to the current account before restricting the ACL, matching _environment_file in tests/native/test_personal_service_lifecycle.py, which already carries this call for the same reason. The two unit fixtures were written alongside it but never ran on Windows CI until now, so they missed it. Refs oceanbase#1627 * test(service): pin the contract that native command output is never dropped The six Windows encoding cases run in CI now, but none of them fails against the pre-fix capture. Hosted runners report an OEM code page of 437, where icacls writes the fixture's non-ASCII path as question marks and every byte decodes, so text=True would have produced the same green result. The one case that does force a decode failure skips on 437 for the same reason. Drive run_windows_command with a child process that writes 0x81, which is undefined in the ANSI code page an English console reports and an incomplete lead byte in a DBCS one. The pre-fix capture therefore loses it on either, leaving returncode 0 with stdout as None, while the OEM decode either succeeds or raises. Asserting that the caller sees a string or a reported failure holds on any code page, so the case needs no skip and turns red on a revert. Refs oceanbase#1627
* feat(pi): promote integration to Full profile * style: apply manifest test formatting
…d adapter conformance suite (oceanbase#1483) * feat(authz): shared filter derivation and snapshot decision boundary - extract resolve_resource_filter derivation shared by both providers (fixes latent parent-type drift between the builtin and Casbin copies) - add RelationalAccessRepository.decision_snapshot: revision, active bindings and ownership read in one transaction; providers use it when offered, else bounded revision-check-and-retry, failing closed when no stable read is obtained (AccessRepository protocol unchanged) - one captured evaluation time per decision for all expiry comparisons - conformance suite: deliberately-interleaving regression for the read-consistency gap, single-transaction pin, fail-closed on unstable revision, point/list agreement and cross-provider filter parity, idempotency ledger conflicts across operations and payloads - no persistent Casbin policy store added (rationale in PR) * Fix type errors in snapshot-boundary tests and decision snapshot row annotations * Annotate decision_snapshot row buffers as Sequence[Mapping[str, Any]] so SQLAlchemy Mapping results type-check. * Make the test doubles nominal subtypes (AsyncDatabase / RelationalAccessRepository) instead of duck-typed wrappers, and hide the decision_snapshot capability via __getattribute__ so the bounded-retry path stays exercised. * Assert policy revision is not None before int() conversion. * Use Mapping[Any, Any] row annotations so ty accepts SQLAlchemy mappings ty infers RowMapping keys as a union type and Mapping is invariant in its key type, so Sequence[Mapping[str, Any]] rejects the rows returned by mappings().all(). Mapping[Any, Any] is accepted by both ty and pyright. * fix(authz): pin the decision read snapshot, fail closed when it moves sqlite3's legacy transaction control does not BEGIN for a bare SELECT, so the transaction decision_snapshot reads through was not a snapshot: a commit landing between the revision read and the bindings read produced a decision labelled with the pre-commit revision but built from post-commit bindings. read_decision_state also delegated on the mere presence of decision_snapshot, leaving the revision-check-and-retry fallback unreachable. - _pin_read_snapshot issues SAVEPOINT rather than BEGIN, so it composes with a caller-owned transaction instead of raising. - decision_snapshot re-reads the revision before releasing the snapshot and raises policy-snapshot-unstable instead of returning the mixture; read_decision_state retries within its existing budget. A READ COMMITTED profile therefore fails closed rather than returning a stale label. - Adds a two-connection interleaving regression against the real snapshot path, and a fail-closed case for a profile that cannot pin isolation. Reported by @Teingi in review. * fix(authz): type the snapshot seams and patch them where they are read The `quality` job failed on `ty check` with three `invalid-assignment` diagnostics: the interleaving tests rebound `_read_policy_revision` and `_pin_read_snapshot` on the *module*, while the attribute being assigned to is a module-level function whose declared signature the local replacement did not match (a different parameter name, and a body that ignores it). That was a symptom of the real problem: the tests were patching a seam the reader does not go through. `decision_snapshot` called the module functions directly, so a test could intercept them without the interception being tied to the code path under test. - Promote both helpers to overridable methods on `RelationalAccessRepository` and call them via `self` from `decision_snapshot`. The module functions stay as the default implementations; the seam is now the instance the reader actually goes through. - Patch that seam with `patch.object` in the two interleaving tests, and drop the module attribute rebinding plus the now-unused `authz_repository` import. `uv run ty check` no longer reports any diagnostic in `tests/test_access_snapshot_boundary.py`; the remaining baseline diagnostics are unchanged (`Found 9` vs `Found 12` before, none in a file this PR touches). `ruff check` / `ruff format --check` clean. --------- Co-authored-by: SanJiu <tlyyxjz@users.noreply.github.com>
* fix(codex): verify native MCP authentication on Windows * fix(codex): address native MCP auth review * fix(codex): validate native MCP bearer header * fix(codex): accept case-insensitive bearer scheme
…ceanbase#1630) * feat(dashboard): add profile reading and exact handoff export * fix(dashboard): preserve reading text and theme recovery * fix(dashboard): preserve indented code in markdown export * fix(dashboard): preserve multiline indented handoff text * fix(dashboard): escape equals lines in handoff export * fix(dashboard): preserve evidence login return
…base#1629) * docs(rfc): design profile dashboard and exact handoff export * docs(rfc): assign RFC 1629 and clarify singleton selection * docs(rfc): remove UI mockup images
oceanbase#1586) * feat: record recurring failures and gate their repair (oceanbase#1557) Add failure signatures, append-only recurrence observations, runtime ledger orchestration, persistence, statistics, API contract updates, and the related documentation and tests. Keep recurrence evidence write-only on the outcome path and expose the required ScopeStats.recurrence block. Constraint: ScopeStats.recurrence is a required breaking API field; Windows-only baseline failures and seven pre-existing ty diagnostics remain outside this change. Tested: 109 focused tests; 48 contract tests; 26 integration-manifest tests; ruff check and format check; generated schema and enabled server OpenAPI match. Not-tested: make check remains non-zero only because of the seven baseline ty diagnostics; full unit test comparison is 2252 passed, 58 skipped, 47 failed on the implementation tree versus 2177 passed, 58 skipped, 46 failed on clean baseline. Co-authored-by: OmX <omx@oh-my-codex.dev> Signed-off-by: Xin.Zh <alexstocks@foxmail.com> * fix: align CI checks with recurring failure API Ignore Mermaid diagrams in license-eye because the checker cannot determine their comment style, and update the existing base-access E2E assertion for the optional Experience failure field. Constraint: Preserve the required ScopeStats.recurrence and ExperienceProposal.failure contract without weakening validation. Tested: targeted base-access E2E test and API contract/JS tests pass; Ruff check and format check pass; licenserc parses as YAML. Co-authored-by: OmX <omx@oh-my-codex.dev> Signed-off-by: Xin.Zh <alexstocks@foxmail.com> * fix: harden recurring failure evidence and replay * fix: reject unresolved handoff revisions * fix: preserve complete recurrence candidate snapshots Lore: A scope-head match must use every active Experience head because its persisted candidate snapshot is evidence for an immutable recurrence decision. Constraint: Keep the independent 64-entry statistics Handoff scan unchanged; this only removes undocumented ledger candidate truncation. Tested: 78 focused recurrence tests and 48 API contract tests pass; Ruff, targeted ty, and git diff checks pass. Not-tested: Full Windows prek remains blocked by nine pre-existing artifact-processing and os.WNOHANG ty diagnostics outside this change. Co-authored-by: OmX <omx@oh-my-codex.dev> * fix(runtime): keep recurrence ledger evidence transactional Lore: SQLite savepoints can commit independently when no real outer BEGIN exists, and recurrence verdicts must be backed by resolvable nested evidence. Constraint: Ledger append savepoints must roll back with the incubation transaction, and terminal recurred/avoided verdicts must not be written when their TaskOutcome item evidence cannot be resolved in the current scope. Tested: uv run pytest -q tests/builtin/persistence/test_recurrence_repository.py tests/builtin/runtime/test_recurrence_ledger.py tests/e2e/test_recurring_failure_repair.py Tested: uv run ruff check src/powercontext/builtin/persistence/recurrence.py src/powercontext/builtin/runtime/recurrence.py src/powercontext/builtin/runtime/relational.py tests/builtin/persistence/test_recurrence_repository.py tests/builtin/runtime/test_recurrence_ledger.py tests/e2e/test_recurring_failure_repair.py Tested: uv run ruff format --check src/powercontext/builtin/persistence/recurrence.py src/powercontext/builtin/runtime/recurrence.py src/powercontext/builtin/runtime/relational.py tests/builtin/persistence/test_recurrence_repository.py tests/builtin/runtime/test_recurrence_ledger.py tests/e2e/test_recurring_failure_repair.py Tested: uv run ty check src/powercontext/builtin/persistence/recurrence.py src/powercontext/builtin/runtime/recurrence.py src/powercontext/builtin/runtime/relational.py tests/builtin/persistence/test_recurrence_repository.py tests/builtin/runtime/test_recurrence_ledger.py tests/e2e/test_recurring_failure_repair.py Co-authored-by: OmX <omx@oh-my-codex.dev> --------- Signed-off-by: Xin.Zh <alexstocks@foxmail.com> Co-authored-by: OmX <omx@oh-my-codex.dev>
…xt (oceanbase#1596) * feat(runtime): add bounded recall sufficiency gate Adds a model-free pre-render sufficiency gate that runs at most two bounded expansion rounds whose only action is lowering the fusion admission floor. It is disabled by default and byte-identical when off, and ships no HTTP, contract or write-path changes. * wip: RFC 1560 observability increment (T01-T03 partial, not green) Intermediate checkpoint of the second increment for RFC 1560 recall_sufficiency_gate: the observability and cost-accounting surface. Committed on a WIP branch so PR oceanbase#1596 (branch feat/recall-gate-impl, 14cbabf) stays at its submitted, green state. Design: .workbuddy/artifacts/rfc1560-observability-design.md Rulings: .workbuddy/artifacts/rfc1560-decisions.md (section 三·补二) Done: AdmissionCounts in artifacts/search.py; RecallBudgetView and the budget probe (PreparedContextBuilder.probe_budget); RecallEffort reshaped to the RFC's fields (assessment, expansion_actions, rounds = 1 + len(...), admission_by_family, added_embeddings, added_generation_calls, and the dropped sub-counts); omissions sub-counts through _fit_entry, _fit_entries, _build_text and the assembly path; ExperienceSearchOutcome and TopicMemoryFusionOutcome; per-family retrieved/admitted counters for all three searchable families; the query_embedding reuse parameter on MemoryService.search; SearchPlan.rerank_candidate_limit and RecallSufficiencyPolicy.round2_rerank_candidate_limit removed per the RFC. NOT done (T04): RecallEffortSink, the _prepare_build -> (build, effort) tuple, removal of PreparedContextBuild.recall_effort, the context.build aggregate counters, and the e2e test updates that depend on them. KNOWN RED: tests/e2e/test_recall_sufficiency_gate.py still reads build.recall_effort and the old rounds/gate_reason semantics, so it has not been adapted and will fail until T04 lands. * feat(runtime): expose recall gate effort through a sink * fix(runtime): address recall gate review feedback Lore: Preserve public recall search contracts and make bounded expansion fail open without evicting round-zero candidates. Constraint: Keep the fix scoped to Copilot's PR oceanbase#1596 recall-gate feedback; leave unrelated artifact_processing ty diagnostics untouched. Tested: uv run pytest tests/builtin/artifacts/topic_memory/test_fusion.py tests/builtin/persistence/test_topic_memory.py::test_topic_memory_search_threads_lowered_fts_floor_into_the_backend tests/builtin/persistence/test_experience_index.py tests/builtin/test_publication.py::test_published_experience_is_searchable_in_target_scope_immediately; uv run pytest tests/builtin/runtime/test_recall_sufficiency.py tests/builtin/runtime/test_prepared_context.py::test_expanded_memory_cap_preserves_the_round_zero_prefix tests/builtin/runtime/test_prepared_context.py::test_expanded_experience_cap_preserves_the_round_zero_prefix tests/builtin/runtime/test_prepared_context.py::test_probe_budget_is_budget_bound_at_the_byte_floor tests/builtin/runtime/test_prepared_context.py::test_probe_budget_is_budget_bound_when_the_fit_drops_items_and_leaves_no_headroom; uv run pytest tests/e2e/test_recall_sufficiency_gate.py; uv run prek run ruff-check -a; uv run prek run ruff-format -a; git diff --check. Not-tested: uv run ty check and uv run prek run -a remain blocked by pre-existing artifact_processing.py and Windows os.WNOHANG ty diagnostics unrelated to this change. Co-authored-by: OmX <omx@oh-my-codex.dev> * test: align OceanBase FTS assertions with fusion admission Lore: FTS indexes must return match candidates so the fusion layer can account for retrieved and admitted Topic Memory results. Constraint: Keep coverage admission out of backend SQL; preserve the public index and query-channel assertions. Tested: OceanBase Topic Memory tests 4 passed, 1 skipped; recall gate and Topic Memory suites 160 passed; Ruff and targeted ty pass. Not-tested: Live OceanBase remains skipped without POWERCONTEXT_TEST_OCEANBASE_URL. Co-authored-by: OmX <omx@oh-my-codex.dev> * fix(runtime): avoid unrecoverable recall expansions Lore: Keep Topic Memory admission accounting separate from backend pool sizing, and keep recall expansion tied to candidates that a lower floor can actually recover. Constraint: Preserve the public Topic Memory search limit and disabled-by-default recall gate behavior; only widen the internal per-channel candidate pool and stop redundant expansion when round zero fully admitted the retrieved family. Tested: uv run pytest -q tests/builtin/persistence/test_topic_memory.py tests/builtin/persistence/test_topic_memory_oceanbase.py tests/builtin/artifacts/topic_memory/test_fusion.py tests/builtin/runtime/test_recall_sufficiency.py tests/builtin/runtime/test_prepared_context.py tests/e2e/test_recall_sufficiency_gate.py; uv run ruff check targeted changed files; uv run ruff format --check targeted changed files; uv run ty check targeted changed files; git diff --check. Co-authored-by: OmX <omx@oh-my-codex.dev> * test: align topic memory request bounds with internal pool Lore: Topic Memory public search limits now live at the repository/API boundary, while request candidate limits describe the internal channel pool used before fusion admission. Constraint: Preserve the public 20-hit search contract; this only updates the model-layer test to assert the internal candidate bound. Tested: uv run pytest -q tests/builtin/artifacts/topic_memory/test_models.py::test_topic_memory_search_request_enforces_query_and_internal_candidate_bounds Tested: uv run pytest -q tests/builtin/artifacts/topic_memory/test_models.py tests/builtin/persistence/test_topic_memory.py tests/builtin/persistence/test_topic_memory_oceanbase.py tests/builtin/artifacts/topic_memory/test_fusion.py tests/builtin/runtime/test_recall_sufficiency.py tests/builtin/runtime/test_prepared_context.py tests/e2e/test_recall_sufficiency_gate.py Tested: uv run ruff check src/powercontext/builtin/artifacts/topic_memory/__init__.py src/powercontext/builtin/artifacts/topic_memory/models.py src/powercontext/builtin/persistence/topic_memory.py src/powercontext/builtin/persistence/sqlite/topic_memory_index.py src/powercontext/builtin/runtime/application.py src/powercontext/builtin/runtime/recall_sufficiency.py tests/builtin/artifacts/topic_memory/test_models.py tests/builtin/persistence/test_topic_memory.py tests/builtin/runtime/test_recall_sufficiency.py tests/e2e/test_recall_sufficiency_gate.py Tested: uv run ruff format --check src/powercontext/builtin/artifacts/topic_memory/__init__.py src/powercontext/builtin/artifacts/topic_memory/models.py src/powercontext/builtin/persistence/topic_memory.py src/powercontext/builtin/persistence/sqlite/topic_memory_index.py src/powercontext/builtin/runtime/application.py src/powercontext/builtin/runtime/recall_sufficiency.py tests/builtin/artifacts/topic_memory/test_models.py tests/builtin/persistence/test_topic_memory.py tests/builtin/runtime/test_recall_sufficiency.py tests/e2e/test_recall_sufficiency_gate.py Tested: uv run ty check src/powercontext/builtin/artifacts/topic_memory/__init__.py src/powercontext/builtin/artifacts/topic_memory/models.py src/powercontext/builtin/persistence/topic_memory.py src/powercontext/builtin/persistence/sqlite/topic_memory_index.py src/powercontext/builtin/runtime/application.py src/powercontext/builtin/runtime/recall_sufficiency.py tests/builtin/artifacts/topic_memory/test_models.py tests/builtin/persistence/test_topic_memory.py tests/builtin/runtime/test_recall_sufficiency.py tests/e2e/test_recall_sufficiency_gate.py Tested: git diff --check Co-authored-by: OmX <omx@oh-my-codex.dev> * fix(runtime): stop unrecoverable recall retries Lore: Recall expansion should only continue while the latest round shows candidates that lowering the floor can still recover. Topic Memory default search also needs to preserve eligible FTS candidates before truncating the internal pool. Constraint: Preserve the public Topic Memory search limit and existing exported fusion helper contract; keep admission counts as aggregate in-process metadata only. Tested: uv run pytest -q tests/builtin/persistence/test_topic_memory.py::test_default_topic_memory_search_preserves_eligible_candidates_before_truncating_the_pool tests/e2e/test_recall_sufficiency_gate.py::test_recoverability_is_refreshed_after_an_expansion_round Tested: uv run pytest -q tests/builtin/persistence/test_topic_memory.py tests/builtin/persistence/test_topic_memory_oceanbase.py tests/builtin/artifacts/topic_memory/test_fusion.py tests/builtin/runtime/test_recall_sufficiency.py tests/builtin/runtime/test_prepared_context.py tests/e2e/test_recall_sufficiency_gate.py Tested: uv run pytest -q tests/builtin/artifacts/topic_memory/test_models.py tests/builtin/artifacts/topic_memory/test_fusion.py tests/builtin/persistence/test_topic_memory.py tests/builtin/persistence/test_topic_memory_oceanbase.py tests/e2e/test_recall_sufficiency_gate.py Tested: uv run ruff check src/powercontext/builtin/artifacts/topic_memory/models.py src/powercontext/builtin/artifacts/topic_memory/fusion.py src/powercontext/builtin/persistence/sqlite/topic_memory_index.py src/powercontext/builtin/persistence/oceanbase/topic_memory_index.py src/powercontext/builtin/persistence/topic_memory_index.py src/powercontext/builtin/runtime/application.py tests/builtin/persistence/test_topic_memory.py tests/builtin/persistence/test_topic_memory_oceanbase.py tests/e2e/test_recall_sufficiency_gate.py Tested: uv run ruff format --check src/powercontext/builtin/artifacts/topic_memory/models.py src/powercontext/builtin/artifacts/topic_memory/fusion.py src/powercontext/builtin/persistence/sqlite/topic_memory_index.py src/powercontext/builtin/persistence/oceanbase/topic_memory_index.py src/powercontext/builtin/persistence/topic_memory_index.py src/powercontext/builtin/runtime/application.py tests/builtin/persistence/test_topic_memory.py tests/builtin/persistence/test_topic_memory_oceanbase.py tests/e2e/test_recall_sufficiency_gate.py Tested: uv run ty check src/powercontext/builtin/artifacts/topic_memory/models.py src/powercontext/builtin/artifacts/topic_memory/fusion.py src/powercontext/builtin/persistence/sqlite/topic_memory_index.py src/powercontext/builtin/persistence/oceanbase/topic_memory_index.py src/powercontext/builtin/persistence/topic_memory_index.py src/powercontext/builtin/runtime/application.py tests/builtin/persistence/test_topic_memory.py tests/builtin/persistence/test_topic_memory_oceanbase.py tests/e2e/test_recall_sufficiency_gate.py Tested: git diff --check Co-authored-by: OmX <omx@oh-my-codex.dev> * fix(runtime): separate topic admission rejection counts Lore: Topic Memory recall traces can keep pre-admission retrieved counts that exceed the capped candidate pool. Constraint: Recall expansion recoverability must only count candidates rejected by admission floors, not rows lost to backend candidate caps. Tested: uv run pytest -q tests/builtin/persistence/test_topic_memory.py tests/builtin/persistence/test_topic_memory_oceanbase.py tests/builtin/artifacts/topic_memory/test_fusion.py tests/builtin/runtime/test_recall_sufficiency.py tests/builtin/runtime/test_prepared_context.py tests/e2e/test_recall_sufficiency_gate.py Tested: uv run ruff check src/powercontext/builtin/artifacts/search.py src/powercontext/builtin/artifacts/topic_memory/models.py src/powercontext/builtin/artifacts/topic_memory/fusion.py src/powercontext/builtin/persistence/sqlite/topic_memory_index.py src/powercontext/builtin/persistence/oceanbase/topic_memory_index.py src/powercontext/builtin/persistence/topic_memory.py src/powercontext/builtin/persistence/topic_memory_index.py src/powercontext/builtin/runtime/application.py tests/builtin/persistence/test_topic_memory_oceanbase.py tests/builtin/runtime/test_recall_sufficiency.py tests/e2e/test_recall_sufficiency_gate.py Tested: uv run ruff format --check src/powercontext/builtin/artifacts/search.py src/powercontext/builtin/artifacts/topic_memory/models.py src/powercontext/builtin/artifacts/topic_memory/fusion.py src/powercontext/builtin/persistence/sqlite/topic_memory_index.py src/powercontext/builtin/persistence/oceanbase/topic_memory_index.py src/powercontext/builtin/persistence/topic_memory.py src/powercontext/builtin/persistence/topic_memory_index.py src/powercontext/builtin/runtime/application.py tests/builtin/persistence/test_topic_memory_oceanbase.py tests/builtin/runtime/test_recall_sufficiency.py tests/e2e/test_recall_sufficiency_gate.py Tested: uv run ty check src/powercontext/builtin/artifacts/search.py src/powercontext/builtin/artifacts/topic_memory/models.py src/powercontext/builtin/artifacts/topic_memory/fusion.py src/powercontext/builtin/persistence/sqlite/topic_memory_index.py src/powercontext/builtin/persistence/oceanbase/topic_memory_index.py src/powercontext/builtin/persistence/topic_memory.py src/powercontext/builtin/persistence/topic_memory_index.py src/powercontext/builtin/runtime/application.py tests/builtin/persistence/test_topic_memory_oceanbase.py tests/builtin/runtime/test_recall_sufficiency.py tests/e2e/test_recall_sufficiency_gate.py Co-authored-by: OmX <omx@oh-my-codex.dev> --------- Co-authored-by: OmX <omx@oh-my-codex.dev>
feat(tags): support all artifact families
* docs(rfc): propose desktop control center * docs(rfc): assign desktop proposal number 1455 * docs(rfc): align desktop design with current contracts * docs(rfc): choose React TypeScript and Vite for desktop * docs(rfc): focus desktop proposal on user model and contracts * ci: rerun checks after macOS service startup timeout * test(native): sample stalled LaunchAgent before cleanup * ci(macos): compile installed bytecode before service lifecycle
Author
|
Closing and withdrawing this patch. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
19 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Completes the review and CI fixes for oceanbase#1652.\n\n- adds the synchronized Chinese RFC\n- renames both RFC files to the assigned 1652 number\n- fixes local RFC links used by the website build\n- updates merged oceanbase#1586/oceanbase#1596 references to RFC 1557/1560\n- clarifies that entry_content_hash includes kind, text, Source refs, and Artifact refs\n- preserves evidence when normalized text matches but evidence refs differ\n\nValidation:\n- make check\n- make docs-test