docs(prompts): Order A3 master prompt draft (bug-class taxonomy + canary recall for Codex) #20

Merged
pdurlej merged 1 commit from claude/orders/order-a3-prompt-draft into main 2026-05-02 02:46:35 +02:00
Collaborator

Summary

Drafted by GLM-5.1 sisyphus (oh-my-opencode framework, ULW-LOOP mode) per .sisyphus/plans/order-a3-prompt.md, following prompts/order-c-codex-askpass.md gold standard. 346 lines covering bug-class taxonomy storage, recall, and convergent-class detection.

Why DRAFT

Pre-Oracle-review draft. Operator pastes prompt to Oracle UI for sanity check before delegation to Codex. Same pattern as Order C prompt PR #11 (Oracle-validated before merge).

Scope (per charter §3 review-by-impact)

Small impact: docs only, single new file (prompts/order-a3-bug-class-taxonomy.md), no behavior change → ensemble review optional. Companion evidence file in .sisyphus/evidence/order-a3-prompt-summary.md.

When Codex implements per this prompt: ~200 LOC code + ~200 LOC tests = medium impact → 3+3 ensemble required at implementation PR.

Key clarification

A3 does NOT add the bug_class field to reviewer schema — that's already documented in charter §3 "Bug-class clustering hints" sub-rule (Oracle Q5 ruling 2026-05-01). A3 operationalizes:

  1. Append-only JSONL storage of emitted bug_class tags
  2. Cross-canary recall (last 10 canaries window)
  3. Per-PR convergent-class detection (≥2 reviewers same class)
  4. Operator alert when class hits ≥3 occurrences in window

Reviewer prompts (_base.py) are NOT touched — schema field already exists.

Linkage

  • Closes (when implementation lands) open loop reviewer-fix-patch-schema step A3 (deadline 2026-05-22)
  • Final of 3 split orders (A1 PR #18, A2 PR #19, A3 this PR)
  • Codex implementation must use linked-worktree pattern per charter §4 v2 (PR #15)

Token-arbitrage

GLM-5.1 wrote 346 lines on flat-fee z.ai subscription. Claude orchestrator reviewed + committed (~5k tokens). Estimated savings: 17-22k orchestrator tokens.

Co-authors

  • GLM-5.1 (z.ai sisyphus draft via OpenCode CLI — opencode run -m zai-coding-plan/glm-5.1)
  • Claude Opus 4.7 (1M context, orchestrator review + commit)
## Summary Drafted by **GLM-5.1 sisyphus** (oh-my-opencode framework, ULW-LOOP mode) per `.sisyphus/plans/order-a3-prompt.md`, following `prompts/order-c-codex-askpass.md` gold standard. 346 lines covering bug-class taxonomy storage, recall, and convergent-class detection. ## Why DRAFT **Pre-Oracle-review draft.** Operator pastes prompt to Oracle UI for sanity check before delegation to Codex. Same pattern as Order C prompt PR #11 (Oracle-validated before merge). ## Scope (per charter §3 review-by-impact) **Small impact**: docs only, single new file (`prompts/order-a3-bug-class-taxonomy.md`), no behavior change → ensemble review optional. Companion evidence file in `.sisyphus/evidence/order-a3-prompt-summary.md`. When Codex implements per this prompt: ~200 LOC code + ~200 LOC tests = **medium impact** → 3+3 ensemble required at implementation PR. ## Key clarification A3 does **NOT** add the `bug_class` field to reviewer schema — that's already documented in charter §3 "Bug-class clustering hints" sub-rule (Oracle Q5 ruling 2026-05-01). A3 operationalizes: 1. Append-only JSONL storage of emitted bug_class tags 2. Cross-canary recall (last 10 canaries window) 3. Per-PR convergent-class detection (≥2 reviewers same class) 4. Operator alert when class hits ≥3 occurrences in window Reviewer prompts (`_base.py`) are NOT touched — schema field already exists. ## Linkage - Closes (when implementation lands) open loop `reviewer-fix-patch-schema` step A3 (deadline 2026-05-22) - Final of 3 split orders (A1 PR #18, A2 PR #19, A3 this PR) - Codex implementation must use linked-worktree pattern per charter §4 v2 (PR #15) ## Token-arbitrage GLM-5.1 wrote 346 lines on flat-fee z.ai subscription. Claude orchestrator reviewed + committed (~5k tokens). Estimated savings: 17-22k orchestrator tokens. ## Co-authors - GLM-5.1 (z.ai sisyphus draft via OpenCode CLI — `opencode run -m zai-coding-plan/glm-5.1`) - Claude Opus 4.7 (1M context, orchestrator review + commit)
Drafted by GLM-5.1 sisyphus per .sisyphus/plans/order-a3-prompt.md, following
prompts/order-c-codex-askpass.md gold standard. 346 lines covering:
- Operationalization of charter §3 "Bug-class clustering hints" (Oracle Q5 ruling 2026-05-01)
- tools/taxonomy.py module spec (~80 LOC) — record_class, recall_classes, convergent_classes
- Append-only JSONL storage at state/reviews/_taxonomy.jsonl
- Convergent-class detection in decision comments (≥2 reviewers same bug_class)
- Cross-canary recall hook with operator alert at ≥3 occurrences in last 10 canaries
- 8+ tests (record + recall + convergent + alert threshold + JSONL durability)
- NO reviewer prompt edits (_base.py untouched per plan hard constraint)
- Bootstrap rule for linked worktree (per charter §4 v2 / Order C-followup)

Pre-Oracle-review draft. Operator pastes to Oracle UI before delegation to Codex.

Per charter §3 review-by-impact: small impact (docs only, single new file,
no behavior change) → ensemble review optional.

Closes (when implementation lands) open loop reviewer-fix-patch-schema step A3
(deadline 2026-05-22). Final of 3 split orders (A1 PR#18, A2 PR#19, A3 this).

Co-authored-by: GLM-5.1 (z.ai sisyphus draft via OpenCode CLI)
Co-authored-by: Claude Opus 4.7 (1M context, orchestrator review + commit)
Sign in to join this conversation.
No reviewers
No labels
W6d-automerge-calibration
agent/claude-code
agent/codex
agent/hermes
agent/iskra
agent/ollama
agent/patchwarden
automerge-candidate
class/security-sensitive
cutover-gate
dependency/blocked
dependency/blocks-others
dependency/cross-repo
dependency/needs-confirmation
domain:agents
domain:ci
domain:docs
domain:forgejo
domain:infra
domain:memory
domain:runtime
domain:signal
domain:ux
flow/architecture
flow/blocked
flow/deployed
flow/done
flow/implementation
flow/intake
flow/maintained
flow/observed
flow/ready
flow/refining
flow/retired
flow/review
iterating
judge/codex-candidate
judge/hermes-candidate
judge/low-confidence
judge/needs-refinement
judge/operator-needed
judge/p0
judge/p1
judge/p2
judge/p3
judge/park
judge/patchwarden-candidate
judge/stale-priority
kind/adr
kind/bug
kind/chore
kind/feature
kind/infra
kind/ops
kind/refactor
kind/research
large-impact
merge/auto
merge/manual
merge/manual-dependency-conflict
merge/manual-failing-tests
merge/manual-merge-conflict
merge/manual-missing-review
merge/manual-operator-preference
merge/manual-red-zone
merge/manual-security-sensitive
merge/manual-unclear-scope
merge/manual-unknown
meta
mode:operator-only
mode:patchwarden-iskra-approved
mode:safe-auto
needs-operator-decision
needs-triage
not-ready
observed/erroring
observed/needs-followup
observed/pending
observed/retire-candidate
observed/unused
observed/used
operator-emotional
owner-attention
phase/02
phase/03
priority:p0
priority:p1
priority:p2
priority:p3
proposed
ready-for-agent
ready-for-operator
recovery
review:claude-reviewed
review:codex-reviewed
review:dziadek-reviewed
review:needs-human
risk/exposure
risk/process
risk/product
risk/runtime
safety:external-write
safety:no-prod-mutation
safety:prod-impact
safety:secret-touch
size/large
size/medium
size/small
size/tiny
size/unknown
source/adr
source/agent-generated
source/manual
source/operator-chat
source/voice-note
status:blocked
status:codex-ready
status:merged:pending-evidence
status:needs-evidence
status:operator-needed
status:parked
tier/full
tier/lite
tier/stacked
tier:0-platform-substrate
tier:1-iskra-value-layer
tier:2-tools-products-modules
type:bug
type:chore
type:docs
type:feat
type:policy
type:research
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
pdurlej/platform!20
No description provided.