tegwick
7cce276d32
Re-ingested under schema v2 (populates error_snippets) and re-ran detect over 27 real sessions. Added a 'content-level root causes' section to docs/ASSESSMENT-infra-friction.md: top recurring error is Edit/Write-before-Read (12/27 sessions, 8 repos), then stale-read conflicts, a cross-flavor (claude+grok) make fix-consistency failure, and State Hub MCP instability. Documented a fingerprint-noise caveat. WP-0006 finished; suite 98/98. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
3.1 KiB
id, type, title, domain, repo, status, owner, topic_slug, created, updated, state_hub_workstream_id
| id | type | title | domain | repo | status | owner | topic_slug | created | updated | state_hub_workstream_id |
|---|---|---|---|---|---|---|---|---|---|---|
| AGENTIC-WP-0006 | workplan | Coding Session Memory — Error-Body Mining (content-level root causes) | helix_forge | agentic-resources | finished | codex | helix-forge | 2026-06-07 | 2026-06-07 | c6e44147-15fd-4cfa-ab2d-87461a6858f1 |
Coding Session Memory — Error-Body Mining
The friction assessment (ASSESSMENT-infra-friction.md) can see that a session was expensive (tool-mix, cost, overhead share) but not always why at the content level — the specific error messages and repeated failed approaches. The digest captures tool histograms and prompt/response snippets, but not error-body text. This workplan closes that gap so Detect can surface recurring root-cause errors, not just coarse markers.
Approach: capture normalized error fingerprints + samples into the durable Tier 2 digest (raw Tier 1 blobs are evictable, so mining must persist into the digest), then cluster recurring fingerprints across sessions into candidate problem patterns through the existing clusterer. No new capture source — this reads the event/blob bodies already ingested.
Capture Error-Body Snippets into the Digest
id: AGENTIC-WP-0006-T01
status: done
priority: high
state_hub_task_id: "136a0a73-61c2-4390-876c-de3880a967e6"
Extend core/digest.py build_digest to extract, from failed events
(kind=error and tool_result bodies matching the existing _FAIL_HINTS), a
normalized fingerprint (strip paths, line numbers, UUIDs, hex) plus a short
sample, stored as digest["error_snippets"] = [{fingerprint, sample, count, tool}].
Same error across a session collapses to one fingerprint with a count. Durable in
Tier 2 (survives Tier 1 eviction). Bump SCHEMA_VERSION. Unit-tested on synthetic
sessions with repeated and varied errors.
Recurring-Error Signal + Clustering
id: AGENTIC-WP-0006-T02
status: done
priority: high
state_hub_task_id: "1a41b6f5-48bc-4080-bd18-94f2186ef566"
Add detect/signals.py sig_recurring_error keyed on the error fingerprint, so
the same error recurring across sessions/repos/flavors clusters into a candidate
problem pattern (locus = fingerprint; magnitude = occurrences). Feeds the existing
clusterer + cross-flavor flagging, so a root-cause error common to multiple flavors
is flagged cross-flavor. Respects the WP-0005 quality filter. Unit-tested on
synthetic digests sharing a fingerprint.
Re-run Live, Extend Friction Assessment with Root Causes
id: AGENTIC-WP-0006-T03
status: done
priority: medium
state_hub_task_id: "bed16d23-3971-4257-b066-d1e639fef150"
Re-ingest (to populate error_snippets — schema bump invalidates old digests) and
re-run detect over the real local sessions. Add a "content-level root causes"
section to ASSESSMENT-infra-friction.md:
top recurring error fingerprints with counts and affected repos/flavors. Full suite
green. After workplan updates, notify the operator to run from ~/state-hub:
make fix-consistency REPO=agentic-resources