Skip to content

Implementation Plan: T5-P6-A10-WP3 — Make NDJSON Vocabulary Drift Visible in Per-Session Diagnostics#4146

Merged
Trecek merged 6 commits into
developfrom
t5-p6-a10-wp3-make-ndjson-vocabulary-drift-visible-in-per-se/4046
Jun 28, 2026
Merged

Implementation Plan: T5-P6-A10-WP3 — Make NDJSON Vocabulary Drift Visible in Per-Session Diagnostics#4146
Trecek merged 6 commits into
developfrom
t5-p6-a10-wp3-make-ndjson-vocabulary-drift-visible-in-per-se/4046

Conversation

@Trecek

@Trecek Trecek commented Jun 28, 2026

Copy link
Copy Markdown
Collaborator

Summary

Wire the existing seen_ndjson_unknown_event_count and seen_ndjson_unknown_item_count counters — currently dead-ending on ClaudeSessionResult — through SkillResult (via a frozen NdjsonDriftOutcome bundle) into flush_session_log, so they appear in summary.json, sessions.jsonl, and (when non-zero) anomalies.jsonl. Add a doctor check (codex_ndjson_drift, Check 35) that scans sessions.jsonl for codex entries with non-zero drift counters and returns WARNING with the affected count.

Implementation Plan

Plan file: /home/talon/projects/autoskillit-runs/impl-20260628-121003-915180/.autoskillit/temp/make-plan/t5_p6_a10_wp3_ndjson_drift_visibility_plan_2026-06-28_121500.md

Closes #4046

🤖 Generated with Claude Code via AutoSkillit

Token Usage Summary

Step Model count uncached output cache_read peak_ctx turns cache_write time
plan* opus[1m] 2 6.6k 30.7k 2.3M 135.4k 81 151.9k 17m 40s
verify* sonnet 1 132 15.3k 829.1k 63.2k 42 44.7k 7m 23s
implement* MiniMax-M3 1 106.1k 18.9k 5.4M 0 137 0 6m 7s
fix* sonnet 1 430 43.8k 5.7M 147.0k 152 128.8k 22m 52s
audit_impl* sonnet 1 817 18.6k 392.4k 71.0k 23 54.5k 7m 48s
prepare_pr* MiniMax-M3 1 46.0k 3.4k 150.1k 0 12 0 58s
compose_pr* MiniMax-M3 1 36.6k 1.3k 140.7k 0 11 0 36s
review_pr* sonnet 1 222 35.5k 1.7M 100.6k 62 82.9k 8m 3s
resolve_review* sonnet 1 189 8.9k 1.3M 67.3k 53 49.4k 5m 18s
Total 197.1k 176.3k 17.8M 147.0k 512.2k 1h 16m

* Step used a non-Anthropic provider; caching behavior may differ.

Token Efficiency

Step LoC Changed cache_read/LoC cache_write/LoC output/LoC
plan 0
verify 0
implement 306 17566.6 0.0 61.9
fix 41 139541.6 3140.9 1067.1
audit_impl 0
prepare_pr 0
compose_pr 0
review_pr 0
resolve_review 3 417473.0 16458.3 2963.0
Total 350 50996.4 1463.4 503.6

Model Usage Breakdown

Model steps uncached output cache_read cache_write time
opus[1m] 1 6.6k 30.7k 2.3M 151.9k 17m 40s
sonnet 5 1.8k 122.0k 9.9M 360.2k 51m 25s
MiniMax-M3 3 188.7k 23.6k 5.7M 0 7m 41s

Trecek added 6 commits June 28, 2026 12:37
Surface the seen_ndjson_unknown_event_count and seen_ndjson_unknown_item_count
counters — previously dead-ending on ClaudeSessionResult — through SkillResult
(via frozen NdjsonDriftOutcome bundle) into flush_session_log, so they appear
in summary.json, sessions.jsonl, and (when non-zero) anomalies.jsonl with
kind=ndjson_drift severity=warning.

Add a doctor check (codex_ndjson_drift, Check 35) that scans sessions.jsonl for
codex entries with non-zero drift counters and returns WARNING with the
affected count.
…ge, sort NdjsonDriftOutcome import, use backend.name for codex filter
…o match sibling test resilience on WSL2 xdist
Replace `assert "2" in result.message` with `assert "2 codex session" in
result.message` to match the actual message format and prevent a false
pass if the count appears elsewhere in the text.
The inline comment restated what the condition already communicates.
Removed per project rules: only add comments when the WHY is non-obvious.
@Trecek Trecek added this pull request to the merge queue Jun 28, 2026
Merged via the queue into develop with commit 49c3db3 Jun 28, 2026
3 checks passed
@Trecek Trecek deleted the t5-p6-a10-wp3-make-ndjson-vocabulary-drift-visible-in-per-se/4046 branch June 28, 2026 20:42
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant