threadkeeper 0.21.0__tar.gz → 0.22.0__tar.gz

{threadkeeper-0.21.0/threadkeeper.egg-info → threadkeeper-0.22.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: threadkeeper
-Version: 0.21.0
+Version: 0.22.0
 Summary: Multi-agent shared brain across Claude Code/Desktop, Codex, Antigravity CLI, Copilot, and VS Code. Cross-session memory, self-improving skill loops, inter-agent signaling — one local MCP server.
 Author: thread-keeper contributors
 License: MIT
@@ -39,7 +39,7 @@ Provides-Extra: encrypted
 Requires-Dist: keyring>=25.0; extra == "encrypted"
 Requires-Dist: sqlcipher3-wheels>=0.5.7; extra == "encrypted"
 Provides-Extra: dev
-Requires-Dist: pytest>=9.1.1; extra == "dev"
+Requires-Dist: pytest<9.2,>=9.1.1; extra == "dev"
 Requires-Dist: pytest-cov>=7.1.0; extra == "dev"
 Requires-Dist: pytest-forked>=1.6; extra == "dev"
 Requires-Dist: pytest-xdist>=3.5; extra == "dev"
@@ -2280,6 +2280,15 @@ It reports these headline groups over a fixed ground-truth set:
   process. It reports a separate score for thread notes, verbatim memory,
   dialectic claims, and lessons or skills.
 
+- **operational recovery comparison** — a second fully synthetic fixture
+  compares dialogue-only evidence with a bounded action/outcome record. Its
+  fixed thresholds require useful recovery links, zero forbidden-string leaks,
+  bounded storage relative to the dialogue baseline, and zero false causal
+  links. The record shape contains only normalized action names, opaque
+  objective/source references, coarse outcomes, error fingerprints, ordering,
+  recovery links, authority, provenance, and expiry. It has no raw arguments,
+  raw output, transcript text, paths, or secret field.
+
 - **outcome classification** — retrieval failure, reasoning failure,
   stale-memory use, removed-memory use, and privacy-policy violation.
 
@@ -2290,6 +2299,17 @@ unavailable when the optional semantic extra is not installed. `--strict`
 requires the versioned fixture thresholds and zero safety violations, making
 the same command suitable for retrieval/model migration gates.
 
+Action/outcome capture is an **offline-only prototype**. The runtime has no
+live producer or export route. Evaluation records expire after 30 days, and
+the retention pass removes them; `tk-forget --selector-type objective <ref>
+--apply` removes an objective's records and invalidates any linked lesson.
+Records sourced from a forgotten dialog/session/thread are removed by the
+same targeted-erasure path. The existing candidate reviewer may see only the
+bounded recovery candidate; a lesson must keep its
+`action_outcome_candidate:<id>` provenance so invalidation hides it. Live use
+requires a later explicit opt-in plus a passing offline comparison; this
+release provides neither.
+
 For DB concurrency, run the reproducible local gate:
 
 ```bash

{threadkeeper-0.21.0 → threadkeeper-0.22.0}/README.md

@@ -2233,6 +2233,15 @@ It reports these headline groups over a fixed ground-truth set:
   process. It reports a separate score for thread notes, verbatim memory,
   dialectic claims, and lessons or skills.
 
+- **operational recovery comparison** — a second fully synthetic fixture
+  compares dialogue-only evidence with a bounded action/outcome record. Its
+  fixed thresholds require useful recovery links, zero forbidden-string leaks,
+  bounded storage relative to the dialogue baseline, and zero false causal
+  links. The record shape contains only normalized action names, opaque
+  objective/source references, coarse outcomes, error fingerprints, ordering,
+  recovery links, authority, provenance, and expiry. It has no raw arguments,
+  raw output, transcript text, paths, or secret field.
+
 - **outcome classification** — retrieval failure, reasoning failure,
   stale-memory use, removed-memory use, and privacy-policy violation.
 
@@ -2243,6 +2252,17 @@ unavailable when the optional semantic extra is not installed. `--strict`
 requires the versioned fixture thresholds and zero safety violations, making
 the same command suitable for retrieval/model migration gates.
 
+Action/outcome capture is an **offline-only prototype**. The runtime has no
+live producer or export route. Evaluation records expire after 30 days, and
+the retention pass removes them; `tk-forget --selector-type objective <ref>
+--apply` removes an objective's records and invalidates any linked lesson.
+Records sourced from a forgotten dialog/session/thread are removed by the
+same targeted-erasure path. The existing candidate reviewer may see only the
+bounded recovery candidate; a lesson must keep its
+`action_outcome_candidate:<id>` provenance so invalidation hides it. Live use
+requires a later explicit opt-in plus a passing offline comparison; this
+release provides neither.
+
 For DB concurrency, run the reproducible local gate:
 
 ```bash

{threadkeeper-0.21.0 → threadkeeper-0.22.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "threadkeeper"
-version = "0.21.0"
+version = "0.22.0"
 description = "Multi-agent shared brain across Claude Code/Desktop, Codex, Antigravity CLI, Copilot, and VS Code. Cross-session memory, self-improving skill loops, inter-agent signaling — one local MCP server."
 requires-python = ">=3.11"
 authors = [{ name = "thread-keeper contributors" }]
@@ -70,7 +70,9 @@ encrypted = [
 # not compose with --forked (collides on the xdist worker tmp dir) and a few
 # tests still race under it. See CONTRIBUTING → Running tests.
 dev = [
-    "pytest>=9.1.1",
+    # pytest 9.2.0 breaks pytest-forked 1.7.5: every forked test exits with
+    # status 3 before reporting. Lift the cap once pytest-forked supports it.
+    "pytest>=9.1.1,<9.2",
     "pytest-cov>=7.1.0",
     "pytest-forked>=1.6",
     "pytest-xdist>=3.5",

threadkeeper-0.22.0/tests/test_action_outcomes.py

@@ -0,0 +1,266 @@
+"""Offline action/outcome memory stays bounded, deletable, and non-live."""
+from __future__ import annotations
+
+import json
+import time
+
+import pytest
+
+
+def _record_pair(conn, action_outcomes, *, objective="objective-42", source_id="dialog-42"):
+    failed = action_outcomes.record_evaluation_event(
+        conn,
+        event_id="ao-failed",
+        objective_ref=objective,
+        action_name="github.pr.create",
+        outcome="failure",
+        error_fingerprint="http.422.validation",
+        sequence=1,
+        source_kind="dialog",
+        source_id=source_id,
+        authority="observed",
+        principal="fixture-runner",
+        channel="offline-evaluation",
+        created_at=100,
+    )
+    recovered = action_outcomes.record_evaluation_event(
+        conn,
+        event_id="ao-recovered",
+        objective_ref=objective,
+        action_name="github.pr.create",
+        outcome="success",
+        sequence=2,
+        recovered_from=failed,
+        source_kind="dialog",
+        source_id=source_id,
+        authority="observed",
+        principal="fixture-runner",
+        channel="offline-evaluation",
+        created_at=101,
+    )
+    return failed, recovered
+
+
+def test_schema_and_validation_exclude_raw_payloads(fresh_mp):
+    from threadkeeper import action_outcomes
+
+    conn = fresh_mp["db"].get_db()
+    columns = {
+        row["name"]
+        for row in conn.execute("PRAGMA table_info(action_outcome_events)")
+    }
+    assert columns == {
+        "id", "objective_ref", "action_name", "outcome", "error_fingerprint",
+        "recovered_from", "sequence", "source_kind", "source_id",
+        "authority_class", "provenance_version", "created_at", "expires_at",
+    }
+
+    with pytest.raises(ValueError, match="bounded identifier"):
+        action_outcomes.record_evaluation_event(
+            conn,
+            event_id="ao-secret",
+            objective_ref="objective-1",
+            action_name="github.pr.create --body raw-output",
+            outcome="success",
+            sequence=1,
+            source_kind="dialog",
+            source_id="dialog-1",
+            authority="observed",
+            principal="fixture-runner",
+            channel="offline-evaluation",
+        )
+    with pytest.raises(RuntimeError, match="offline evaluation"):
+        action_outcomes.record_live_event()
+
+
+def test_recovery_order_and_authorized_candidate_are_bounded(fresh_mp):
+    from threadkeeper import action_outcomes
+
+    conn = fresh_mp["db"].get_db()
+    failed, recovered = _record_pair(conn, action_outcomes)
+    candidate_id = action_outcomes.derive_recovery_candidate(conn, recovered)
+
+    row = conn.execute(
+        "SELECT content, rationale, source_uuid FROM extract_candidates WHERE id=?",
+        (candidate_id,),
+    ).fetchone()
+    assert row["rationale"] == "operational_recovery"
+    assert row["source_uuid"] == f"action-outcome:{recovered}"
+    assert row["content"] == (
+        "Operational recovery: retry github.pr.create after "
+        "http.422.validation."
+    )
+    assert action_outcomes.candidate_source(conn, candidate_id) == (
+        "action_outcome_candidate", str(candidate_id)
+    )
+    exported = action_outcomes.export_events(conn)
+    assert [row["id"] for row in exported] == [failed, recovered]
+    assert set(exported[0]) == {
+        "id", "objective_ref", "action_name", "outcome", "error_fingerprint",
+        "recovered_from", "sequence", "source_kind", "source_id",
+        "authority_class", "provenance_version", "created_at", "expires_at",
+    }
+
+    with pytest.raises(ValueError, match="earlier failure"):
+        action_outcomes.record_evaluation_event(
+            conn,
+            event_id="ao-out-of-order",
+            objective_ref="objective-42",
+            action_name="github.pr.create",
+            outcome="success",
+            sequence=1,
+            recovered_from=failed,
+            source_kind="dialog",
+            source_id="dialog-42",
+            authority="observed",
+            principal="fixture-runner",
+            channel="offline-evaluation",
+        )
+    with pytest.raises(ValueError, match="same objective"):
+        action_outcomes.record_evaluation_event(
+            conn,
+            event_id="ao-cross-objective",
+            objective_ref="objective-43",
+            action_name="github.pr.create",
+            outcome="success",
+            sequence=2,
+            recovered_from=failed,
+            source_kind="dialog",
+            source_id="dialog-43",
+            authority="observed",
+            principal="fixture-runner",
+            channel="offline-evaluation",
+        )
+
+
+def test_expiry_and_source_or_objective_forgetting_remove_candidates_and_invalidate_lessons(
+    fresh_mp,
+):
+    from threadkeeper import action_outcomes
+    from threadkeeper.authority import source_is_live
+    from threadkeeper.tools.lessons import lesson_append
+
+    conn = fresh_mp["db"].get_db()
+    _failed, recovered = _record_pair(conn, action_outcomes)
+    candidate_id = action_outcomes.derive_recovery_candidate(conn, recovered)
+    conn.commit()
+
+    result = lesson_append(
+        title="Recover GitHub validation errors",
+        summary="Use the linked recovery evidence.",
+        body="Retry the normalized action only after the stable validation error.",
+        source=f"action_outcome_candidate:{candidate_id}",
+    )
+    assert result.startswith("ok slug=recover-github-validation-errors")
+    assert source_is_live(conn, ("lesson", "recover-github-validation-errors"))
+
+    deleted = action_outcomes.forget_objective(conn, "objective-42")
+    assert deleted == {"events": 2, "candidates": 1}
+    assert not source_is_live(conn, ("lesson", "recover-github-validation-errors"))
+    assert conn.execute("SELECT COUNT(*) FROM action_outcome_events").fetchone()[0] == 0
+    assert conn.execute("SELECT COUNT(*) FROM extract_candidates").fetchone()[0] == 0
+
+    old = action_outcomes.record_evaluation_event(
+        conn,
+        event_id="ao-expired",
+        objective_ref="objective-44",
+        action_name="github.pr.create",
+        outcome="failure",
+        error_fingerprint="http.429.rate-limit",
+        sequence=1,
+        source_kind="dialog",
+        source_id="dialog-44",
+        authority="observed",
+        principal="fixture-runner",
+        channel="offline-evaluation",
+        created_at=int(time.time()) - 3 * 86400,
+        retention_days=1,
+    )
+    assert old == "ao-expired"
+    assert action_outcomes.prune_expired(conn, int(time.time())) == {"events": 1, "candidates": 0}
+
+    action_outcomes.record_evaluation_event(
+        conn,
+        event_id="ao-source-forget",
+        objective_ref="objective-45",
+        action_name="github.pr.create",
+        outcome="failure",
+        error_fingerprint="http.429.rate-limit",
+        sequence=1,
+        source_kind="dialog",
+        source_id="dialog-45",
+        authority="observed",
+        principal="fixture-runner",
+        channel="offline-evaluation",
+    )
+    assert action_outcomes.forget_sources(conn, [("dialog", "dialog-45")]) == {
+        "events": 1,
+        "candidates": 0,
+    }
+
+    conn.execute(
+        "INSERT INTO dialog_messages "
+        "(uuid, source, role, content, created_at) VALUES "
+        "('dialog-46', 'pytest', 'user', 'erasable source', 100)",
+    )
+    action_outcomes.record_evaluation_event(
+        conn,
+        event_id="ao-forget-selector",
+        objective_ref="objective-46",
+        action_name="github.pr.create",
+        outcome="failure",
+        error_fingerprint="http.429.rate-limit",
+        sequence=1,
+        source_kind="dialog",
+        source_id="dialog-46",
+        authority="observed",
+        principal="fixture-runner",
+        channel="offline-evaluation",
+    )
+    conn.commit()
+    from threadkeeper.forget import forget_selector
+
+    assert "action_outcomes=1" in forget_selector(
+        "dialog-46", selector_type="uuid", conn=conn,
+    )
+    forget_selector("dialog-46", selector_type="uuid", dry_run=False, conn=conn)
+    assert conn.execute("SELECT COUNT(*) FROM action_outcome_events").fetchone()[0] == 0
+
+
+def test_offline_comparison_sets_thresholds_without_enabling_live_ingest():
+    from threadkeeper import action_outcomes
+
+    fixture = {
+        "dialogue_only": [
+            "tool arguments very-sensitive-arguments tool output unbounded-output " * 20,
+        ],
+        "events": [
+            {
+                "id": "failure", "objective_ref": "objective-1",
+                "action_name": "github.pr.create", "outcome": "failure",
+                "error_fingerprint": "http.422.validation", "sequence": 1,
+            },
+            {
+                "id": "recovery", "objective_ref": "objective-1",
+                "action_name": "github.pr.create", "outcome": "success",
+                "sequence": 2, "recovered_from": "failure",
+            },
+        ],
+        "expected_recoveries": [["failure", "recovery"]],
+        "forbidden_substrings": ["very-sensitive-arguments", "unbounded-output"],
+        "thresholds": {
+            "min_lesson_usefulness": 1.0,
+            "max_privacy_leakage": 0,
+            "max_storage_ratio": 0.5,
+            "max_false_causal_links": 0,
+        },
+    }
+    report = action_outcomes.evaluate_offline_fixture(fixture)
+
+    assert report["thresholds_pass"] is True
+    assert report["bounded_action_outcomes"]["privacy_leakage"] == 0
+    assert report["dialogue_only"]["privacy_leakage"] == 2
+    assert report["bounded_action_outcomes"]["lesson_usefulness"] == 1.0
+    assert report["live_ingestion_allowed"] is False
+    assert action_outcomes.live_ingestion_allowed(True, report) is True
+    assert json.dumps(report)

{threadkeeper-0.21.0 → threadkeeper-0.22.0}/tests/test_memory_eval.py

@@ -166,6 +166,8 @@ def test_smoke_emits_headline_metrics(smoke_report):
     assert r["evidence_recall"]["rate"] == 1.0
     assert r["final_answer"]["rate"] == 1.0
     assert r["thresholds_pass"] is True
+    assert r["operational_recovery"]["thresholds_pass"] is True
+    assert r["operational_recovery"]["live_ingestion_allowed"] is False
 
 
 def test_smoke_covers_every_versioned_axis(smoke_report):

{threadkeeper-0.21.0 → threadkeeper-0.22.0}/tests/test_retention.py

@@ -229,6 +229,35 @@ def test_prune_then_vacuum_checkpoint_round_trip(fresh_mp, monkeypatch):
     assert _count(conn, "notes", "content='after prune'") == 1
 
 
+def test_retention_expires_bounded_action_outcomes(fresh_mp, monkeypatch):
+    from threadkeeper import action_outcomes, retention
+
+    conn = fresh_mp["db"].get_db()
+    action_outcomes.record_evaluation_event(
+        conn,
+        event_id="ao-retention",
+        objective_ref="objective-retention",
+        action_name="github.pr.create",
+        outcome="failure",
+        error_fingerprint="http.429.rate-limit",
+        sequence=1,
+        source_kind="dialog",
+        source_id="dialog-retention",
+        authority="observed",
+        principal="fixture-runner",
+        channel="offline-evaluation",
+        created_at=int(time.time()) - 3 * 86400,
+        retention_days=1,
+    )
+    conn.commit()
+
+    _retention_config(monkeypatch, retention)
+    out = retention.run_retention_pass(force=True)
+
+    assert "action_outcomes=1" in out
+    assert _count(conn, "action_outcome_events") == 0
+
+
 def test_vacuum_rebases_dialog_cursors_and_rebuilds_fts(fresh_mp, monkeypatch):
     """VACUUM may renumber dialog_messages' implicit rowids once dialog rows
     were pruned. The retention vacuum must rebase the rowid ingest cursors to