[codex] add exact failure pattern mining

AGENTS.md

@@ -72,6 +72,7 @@ uvx --from radon radon cc -s -a src tests plugins/openflywheel/scripts/mcp_serve
 python3 ~/.codex/skills/.system/skill-creator/scripts/quick_validate.py plugins/openflywheel/skills/trace-query-planner
 python3 ~/.codex/skills/.system/skill-creator/scripts/quick_validate.py plugins/openflywheel/skills/outcome-recorder
 python3 ~/.codex/skills/.system/skill-creator/scripts/quick_validate.py plugins/openflywheel/skills/failure-miner
+python3 ~/.codex/skills/.system/skill-creator/scripts/quick_validate.py plugins/openflywheel/skills/failure-pattern-miner
 python3 ~/.codex/skills/.system/plugin-creator/scripts/validate_plugin.py plugins/openflywheel
 ```
 

plugins/openflywheel/.codex-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "openflywheel",
-  "version": "0.5.0",
-  "description": "Initialize ITSM-bench harness workspaces, query Langfuse trajectories, and record outcomes and compact failure diagnoses.",
+  "version": "0.6.0",
+  "description": "Initialize ITSM-bench harness workspaces, query Langfuse trajectories, and mine compact failure diagnoses and exact patterns.",
   "author": {
     "name": "OpenFlyWheel"
   },
@@ -10,8 +10,8 @@
   "skills": "./skills/",
   "interface": {
     "displayName": "OpenFlyWheel",
-    "shortDescription": "Prepare ITSM workspaces and mine failures",
-    "longDescription": "Initialize an ITSM-bench agent-harness optimization workspace, inspect bounded Langfuse trajectory evidence, record authoritative outcomes, and persist compact failure diagnoses in the prepared local workspace.",
+    "shortDescription": "Diagnose and mine ITSM failure patterns",
+    "longDescription": "Initialize an ITSM-bench agent-harness optimization workspace, inspect bounded Langfuse evidence, record authoritative outcomes and compact diagnoses, and mine exact recurring patterns without copying trace payloads.",
     "developerName": "OpenFlyWheel",
     "category": "Productivity",
     "capabilities": ["Read", "Write"],

plugins/openflywheel/.mcp.json

@@ -4,7 +4,7 @@
       "command": "uvx",
       "args": [
         "--from",
-        "git+https://github.com/divo12/OpenFlyWheel.git@0d4af6e0499c54765e2f7784b48b53baf96d1ad5",
+        "git+https://github.com/divo12/OpenFlyWheel.git@ab0ef62cbe1e6cddf0bfd8ec61374d10120c61aa",
         "--with",
         "mcp>=1.13,<2",
         "openflywheel-mcp"

plugins/openflywheel/program_templates/itsm.md

@@ -44,6 +44,11 @@ under `.workspace/failures/` before forming a harness hypothesis. Record an inco
 diagnosis when the evidence is insufficient. Do not copy Langfuse trace payloads into the
 workspace; Langfuse remains their source of truth.
 
+After recording the diagnoses for one bounded run, use `$failure-pattern-miner` and call
+`mine_failure_patterns` with only the returned artifact IDs from that run. It groups by failure
+type and exact normalized root cause; results are not semantic clusters. Keep inconclusive
+diagnoses separate, and reread each supporting diagnosis before forming a shared hypothesis.
+
 An intermediate tool error is evidence, not an outcome failure, when the agent recovered
 and the verifier passed. A technically clean trajectory is still a failure when the ITSM
 verifier shows that the required environment state was not achieved.

plugins/openflywheel/skills/failure-pattern-miner/SKILL.md

@@ -0,0 +1,28 @@
+---
+name: failure-pattern-miner
+description: Group explicit compact OpenFlywheel failure diagnoses by issue type and exact normalized root cause. Use after failure-miner records diagnoses for a bounded comparison scope; do not use to diagnose traces, infer outcomes, create semantic clusters, or edit a harness.
+---
+
+# Failure Pattern Miner
+
+Mine repeated patterns from diagnoses already recorded by `$failure-miner`. Use only
+artifact IDs returned by `record_failure` from the same prepared workspace and the same
+comparison scope, such as one baseline or candidate run. Pass one to fifty unique artifact
+IDs to `mine_failure_patterns`; never scan the workspace or substitute trace IDs and paths.
+
+The tool groups supported diagnoses by failure type plus exact normalized root cause. Its
+normalizer masks volatile absolute paths, recognizable opaque identifiers, and numbers before
+fingerprinting the complete normalized cause; the returned cause text is only a bounded display
+excerpt. Results are deterministic exact matches, not semantic clusters: similar wording may
+remain separate, and matching wording does not prove one repair will fix every occurrence.
+Inconclusive diagnoses remain separate and must not be forced into a pattern.
+
+Read patterns in their declared order: occurrence count descending, distinct task count
+descending, latest occurrence descending, then fingerprint ascending. Preserve the returned
+fingerprints, normalized causes, task IDs, trace IDs, artifact IDs, and time bounds. A
+repeated pattern may prioritize later hypothesis work, but reread its compact diagnosis
+artifacts before proposing a harness change.
+
+Do not query Langfuse, call `record_outcome` or `record_failure`, modify diagnosis artifacts,
+merge results from unrelated experiment scopes, generate embeddings, infer a broader cause,
+or edit the harness while following this skill.

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "openflywheel"
-version = "0.5.0"
+version = "0.6.0"
 description = "A governed self-improving agent harness"
 requires-python = ">=3.11"
 dependencies = [

src/ofw/__init__.py

@@ -31,8 +31,15 @@
     FailureDiagnosisError,
     FailureErrorCode,
     FailureEvidenceStatus,
+    FailurePatternMiningError,
+    FailurePatternMiningErrorCode,
+    FailurePatternMiningObservation,
+    FailurePatternMiningStatus,
+    FailurePatternOrdering,
+    FailurePatternSummary,
     FailureType,
     LangfuseOutcomeStore,
+    MineFailurePatternsInput,
     OutcomeErrorCode,
     OutcomeEvaluation,
     OutcomeEvaluationError,
@@ -109,6 +116,12 @@ def editable(self, path: Path) -> EditableFile:
     "FailureDiagnosisError",
     "FailureErrorCode",
     "FailureEvidenceStatus",
+    "FailurePatternMiningError",
+    "FailurePatternMiningErrorCode",
+    "FailurePatternMiningObservation",
+    "FailurePatternMiningStatus",
+    "FailurePatternOrdering",
+    "FailurePatternSummary",
     "FailureType",
     "GitCommit",
     "Harness",
@@ -124,6 +137,7 @@ def editable(self, path: Path) -> EditableFile:
     "LangfuseProject",
     "LangfuseSpan",
     "ModelFingerprint",
+    "MineFailurePatternsInput",
     "OutcomeErrorCode",
     "OutcomeEvaluation",
     "OutcomeEvaluationError",

src/ofw/evaluation/__init__.py

@@ -7,6 +7,15 @@
     FailureEvidenceStatus,
     FailureType,
 )
+from ofw.evaluation.failure_patterns import (
+    FailurePatternMiningError,
+    FailurePatternMiningErrorCode,
+    FailurePatternMiningObservation,
+    FailurePatternMiningStatus,
+    FailurePatternOrdering,
+    FailurePatternSummary,
+    MineFailurePatternsInput,
+)
 from ofw.evaluation.langfuse import (
     LangfuseOutcomeStore,
     OutcomeScoreSubmission,
@@ -26,8 +35,15 @@
     "FailureDiagnosisError",
     "FailureErrorCode",
     "FailureEvidenceStatus",
+    "FailurePatternMiningError",
+    "FailurePatternMiningErrorCode",
+    "FailurePatternMiningObservation",
+    "FailurePatternMiningStatus",
+    "FailurePatternOrdering",
+    "FailurePatternSummary",
     "FailureType",
     "LangfuseOutcomeStore",
+    "MineFailurePatternsInput",
     "OutcomeErrorCode",
     "OutcomeEvaluation",
     "OutcomeEvaluationError",