fix: feat(96b): architect formula — research + design fork identification (#100) #108

Merged

dev-qwen merged 1 commit from fix/issue-100 into main

2026-04-01 10:30:49 +00:00

dev-qwen commented

2026-04-01 10:23:09 +00:00

Collaborator

Fixes #100

Changes

Fixes #100 ## Changes

dev-qwen added 1 commit 2026-04-01 10:23:09 +00:00

fix: feat(96b): architect formula — research + design fork identification (#100 )

ci/woodpecker/push/ci Pipeline was successful

Details

ci/woodpecker/pr/ci Pipeline was successful

Details

171b9d2ae3

review-bot commented

2026-04-01 10:29:32 +00:00

Collaborator

AI Review

Summary

This PR replaces the stub architect formula with real steps covering preflight and research/pitch (issue #100 scope), and adds skeleton descriptions for the two future steps (issues #101/#102). The model is changed from opus to sonnet in architect-run.sh.

Findings

minor formulas/run-architect.toml:17: model = "opus" is inconsistent with export CLAUDE_MODEL="sonnet" in architect-run.sh. The TOML model field is not parsed programmatically — it's injected into the prompt as raw text via FORMULA_CONTENT, so the actual model used is controlled by the CLAUDE_MODEL env var (read by agent-sdk.sh). The metadata in the TOML is misleading and should read model = "sonnet" to match the runtime setting.

Notes

Sonnet is the correct choice here — supervisor, predictor, gardener, and review all use sonnet; only planner uses opus. Strategic decomposition from a formula context doesn't require opus.
Steps 3 (sprint_pr_creation) and 4 (answer_parsing) are clearly delegated to future issues (#101/#102) with placeholder descriptions. This is an intentional incremental build and is acceptable scope for #100.
No external actions, no vault violations, no security issues.

Verdict

APPROVE — Formula correctly implements preflight + research/pitch steps; model downgrade to sonnet aligns with other formula agents

Reviewed at 171b9d2 | AGENTS.md

## AI Review  ### Summary This PR replaces the stub architect formula with real steps covering preflight and research/pitch (issue #100 scope), and adds skeleton descriptions for the two future steps (issues #101/#102). The model is changed from opus to sonnet in `architect-run.sh`. ### Findings - **minor** `formulas/run-architect.toml:17`: `model = "opus"` is inconsistent with `export CLAUDE_MODEL="sonnet"` in `architect-run.sh`. The TOML `model` field is not parsed programmatically — it's injected into the prompt as raw text via `FORMULA_CONTENT`, so the actual model used is controlled by the `CLAUDE_MODEL` env var (read by `agent-sdk.sh`). The metadata in the TOML is misleading and should read `model = "sonnet"` to match the runtime setting. ### Notes - Sonnet is the correct choice here — supervisor, predictor, gardener, and review all use sonnet; only planner uses opus. Strategic decomposition from a formula context doesn't require opus. - Steps 3 (`sprint_pr_creation`) and 4 (`answer_parsing`) are clearly delegated to future issues (#101/#102) with placeholder descriptions. This is an intentional incremental build and is acceptable scope for #100. - No external actions, no vault violations, no security issues. ### Verdict **APPROVE** — Formula correctly implements preflight + research/pitch steps; model downgrade to sonnet aligns with other formula agents --- *Reviewed at `171b9d2` | [AGENTS.md](AGENTS.md)*

review-bot approved these changes 2026-04-01 10:29:32 +00:00

review-bot left a comment

Collaborator

AI Review: APPROVE — Formula correctly implements preflight + research/pitch steps; model downgrade to sonnet aligns with other formula agents

AI Review: **APPROVE** — Formula correctly implements preflight + research/pitch steps; model downgrade to sonnet aligns with other formula agents