docs(overhaul): switch remaining agents to OpenAI
This commit is contained in:
parent
185590e016
commit
e2f5385037
2 changed files with 16 additions and 9 deletions
|
|
@ -161,13 +161,18 @@ features; removing the user-requested Extended particle-range option.
|
|||
The v1 failure was one agent given the whole campaign for eighteen hours. v2
|
||||
keeps the lead in the loop at every chunk.
|
||||
|
||||
**User override 2026-09-04:** all remaining campaign agents use OpenAI
|
||||
models. Do not invoke Claude Code, Anthropic models, or Anthropic credentials.
|
||||
Historical Sonnet/Opus labels elsewhere in this ledger describe already-run
|
||||
work only; they are not instructions for future dispatches.
|
||||
|
||||
| Role | Who | Does | Never |
|
||||
|---|---|---|---|
|
||||
| Lead / architect | the main session (Fable) | writes each chunk contract (files, retail functions + addresses, acceptance tests); dispatches one bounded chunk; verifies every claim against source; runs gates; commits; merges main in at slice boundaries | hands off "complete the slice" |
|
||||
| Implementer | Sonnet, one dispatch per chunk | one file cluster + its tests; build and test green is the return condition; returns a diff summary and open questions | commits; adds evidence infrastructure; starts a third fix round; touches files outside its contract |
|
||||
| Retail reviewer | Opus, one pass per slice on the production port | checks each ported branch against the named decomp with addresses; findings as file:line + decomp citation | reviews tooling |
|
||||
| Architecture reviewer | Opus, only at the two cutover slices (S2, S5) | one owner per fact; layering; lifetime; performance shape | |
|
||||
| Finding verification | the lead, against the source | each reviewer finding is checked by the lead against the decomp/contract/code before it is acted on; at most one Opus skeptic, and only for a blocking finding the lead cannot settle | spawns a skeptic per finding (the S1 review fanned out to ~30 agents this way on 2026-09-02; never again) |
|
||||
| Lead / architect | the main OpenAI Codex session | writes each chunk contract (files, retail functions + addresses, acceptance tests); dispatches one bounded chunk; verifies every claim against source; runs gates; commits; merges main in at slice boundaries | hands off "complete the slice" |
|
||||
| Implementer | one OpenAI implementation agent per chunk | one file cluster + its tests; build and test green is the return condition; returns a diff summary and open questions | commits; adds evidence infrastructure; starts a third fix round; touches files outside its contract |
|
||||
| Retail reviewer | one independent OpenAI review agent per slice on the production port | checks each ported branch against the named decomp with addresses; findings as file:line + decomp citation | reviews tooling |
|
||||
| Architecture reviewer | one independent OpenAI review agent, only at the two cutover slices (S2, S5) | one owner per fact; layering; lifetime; performance shape | |
|
||||
| Finding verification | the lead, against the source | each reviewer finding is checked by the lead against the decomp/contract/code before it is acted on; at most one independent OpenAI skeptic, and only for a blocking finding the lead cannot settle | spawns a skeptic per finding (the S1 review fanned out to ~30 agents this way on 2026-09-02; never again) |
|
||||
|
||||
**Agent budget per slice:** implementers as needed (one per chunk, serial for
|
||||
coupled files); reviewers two (retail lens, and the architecture lens only at
|
||||
|
|
@ -682,7 +687,7 @@ Update immediately when a slice changes state. Chat is not the ledger.
|
|||
| S4-c3b | **LANDED 2026-09-04** — implementation `26e97ba41`, provenance contract `5110bf676`, packet-only correction `3f2f00c9f`; packet §22–§25. Dead classic-group `LocalSortCenters`/`CachedBatch.LocalSortCenter` storage and the alpha camera-parameter/digest chain are deleted. Live per-cell/particle CYpt keys, opaque `SortDistance`, building/private/portal distances, two FIFO lists, all state/barriers, and AP/AD rows remain. | **G3 LEAD SELF-GATE PROVISIONAL PASS; owner acceptance/owner-only rows pending. G4 UNPASSED.** | Retail/deletion lens PASS. Production lens found no code defect and one artifact-provenance omission; packet-only fix round 1 passed its narrow re-review. Fresh campaign: Release 0W/0E; App 132/132; Core 29/29; allocation 2/2 at 0 B; shader 32/32. G3: route 1 13/13, route 2 4/4, route 3 retry 3/3, all exit 0/graceful. The first route-3 attempt stopped before Nanto on registered #462. Exact PNG paths: packet §25. |
|
||||
| S5-c1 | **LANDED 2026-09-04 after the owner-authorized documentation/evidence exception.** Mandatory stop `efb075619` remains in history; final campaign tip `1718832e2`, reviewed scratch tip `b2d56f9e3`. | G4 UNPASSED | Narrow exception re-review PASS. Clean evidence: hermetic 16,760/16,760; inclusive InstalledDat 385 pass / the same documented 10 fail / 1 skip; both exact `d6592d3ac`, `WorktreeDirty=false`, manifests fully verified. Fresh campaign Release 0W/0E, Core 111/111, visibility 38/38, allocation/AP-116 4/4. Lead graphical portal-haze gate PROVISIONAL PASS, exit 0/graceful, 10 PNGs under `logs/selfgate-20260904-171137-s5c1-particle-visibility`; owner inspection pending. |
|
||||
| S5-c2 | **LANDED + REVIEW-CLOSED 2026-09-04; lead graphical gate PROVISIONAL PASS.** Campaign stack `a4de2efc4` -> `048027e71` -> `ef819eedf`; reviewed scratch tip `521f5edda54`; packet §§14–16. The opt-in IA-24 path borrows the exact prior-completed landscape set/flag, selects world casters through S2 CELLARRAY or building `EffectCellId`, selects authored terrain slots, and projects exact active instance runs without rebuilding retained topology. Pack-off remains unchanged. | G4 UNPASSED | Retail/deviation PASS; production's documentation-only fix round 1 narrowly re-reviewed PASS. Fresh campaign Release 0W/0E and affected 137/137. Official evidence: hermetic 16,768/16,768; broader InstalledDat 468 pass / 10 fail / 1 skip; canonical 385 pass / the same 10 fail / 1 skip; manifests and nonpassing identities exact. Valid lead gate `logs/selfgate-20260904-183925-s5c2-candidate-shadow-ab`: seven PNGs, exit 0, graceful logout, no fatal match, no client left. Exact parent `5c106bcdf` A/B at `s5-c2-ab-parent/logs/selfgate-20260904-183815-s5c2-parent-shadow-ab` proves the gate-visible dark lighting switch predates c2; #469/IA-24 carry it for resolution or explicit owner acceptance before G4. Owner PNG inspection pending. |
|
||||
| S5-c3 | **CONTRACTED 2026-09-04; implementation pending.** Packet §17 is binding. | G4 UNPASSED | Exact building part-0 degrade selection, selected-Gfx complete-body gate, selected shell submission, shared retail FPS/automatic-degrade owner, and the three live Config preferences. Sonnet implements in a detached scratch worktree; lead verification and both sequential reviews precede landing. |
|
||||
| S5-c3 | **CONTRACTED 2026-09-04; implementation pending.** Packet §17 is binding. | G4 UNPASSED | Exact building part-0 degrade selection, selected-Gfx complete-body gate, selected shell submission, shared retail FPS/automatic-degrade owner, and the three live Config preferences. Per the owner's 2026-09-04 override, one OpenAI implementation agent works in the detached scratch worktree; lead verification and both sequential OpenAI reviews precede landing. |
|
||||
| S5 | IN FLIGHT — c1 and c2 landed/review-closed/provisionally self-gated; c3 contracted; c4–c5, #469 disposition, closeout, and G4 remain | G4 | Packet §§7 and 12–17 are binding; S5-c3 implementation is next; never merge main before G4. |
|
||||
|
||||
---
|
||||
|
|
|
|||
|
|
@ -6,9 +6,11 @@ S5-c3 CONTRACTED 2026-09-04, implementation pending.
|
|||
**Gate:** G4 remains unpassed. Nothing merges to `main` before G4.
|
||||
|
||||
This packet decomposes plan S5 into bounded chunks. The lead writes each
|
||||
contract, a Sonnet implementer changes only that contract, the lead verifies
|
||||
every retail claim against the named pseudo-C and the paired executable/PDB,
|
||||
and the sequential review lenses run before landing. A chunk that needs a
|
||||
contract, one bounded OpenAI implementation agent changes only that contract,
|
||||
the lead verifies every retail claim against the named pseudo-C and the paired
|
||||
executable/PDB, and independent sequential OpenAI review lenses run before
|
||||
landing. Per the owner's 2026-09-04 override, no remaining campaign step uses
|
||||
Claude Code, Anthropic models, or Anthropic credentials. A chunk that needs a
|
||||
third fix round stops and is written up. Every surviving deviation is entered
|
||||
or corrected in the divergence register in the same implementation commit.
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue