Skip to content

Resolve plan ambiguities, add review tooling and review findings - #134

Closed
kingryad wants to merge 1 commit into
ed-donner:mainfrom
kingryad:plan-revision-and-review
Closed

Resolve plan ambiguities, add review tooling and review findings#134
kingryad wants to merge 1 commit into
ed-donner:mainfrom
kingryad:plan-revision-and-review

Conversation

@kingryad

@kingryad kingryad commented Aug 3, 2026

Copy link
Copy Markdown

Summary

Turns the parts of PLAN.md where the Frontend and Backend agents would each have invented an incompatible answer into explicit contracts, and adds the review tooling that produced the revision.

  • planning/PLAN.md — pins down the SSE frame shape (documented verbatim from the shipped stream.py), a single error envelope with stable codes, the chat actions schema, the priced-set union rule (watchlist ∪ non-zero positions), and trade execution rules. A resolution log records what was decided and why, so the reasoning is recoverable without re-litigating it.
  • .claude/skills/cerebras/SKILL.md — renames the frontmatter cerebras-inferencecerebras. Before this, §9's reference didn't resolve to the skill.
  • .claude/agents/reviewer.md, .claude/commands/doc-review.md — previously untracked, so the workflow that produced this revision wasn't reproducible from a fresh clone.
  • planning/REVIEW.md — current review findings.

Open questions

Four questions for the project owner are recorded at the end of PLAN.md. Q1 blocks frontend work: adding previous_close means editing the market-data subsystem that's marked complete/tested/reviewed. The alternative is dropping the daily change % column.

Findings worth acting on before the next build phase

From REVIEW.md, highest severity first:

  1. backend/app/market/stream.py:17 creates a module-level APIRouter, so create_stream_router() isn't the globals-free factory its docstring claims — two calls register two routes on the same object and the second app silently streams the first app's PriceCache.
  2. previous_close doesn't exist yet (models.py:13-16), and §10 tells the frontend to compute daily change % from it.
  3. .gitignore:17 is lib/ with no leading slash, which silently ignores frontend/lib/ and frontend/src/lib/ — the default shadcn/Tailwind location for the exact stack §10 specifies.
  4. README.md contradicts the revised plan on four points (lazy init, named volume, required API key, deleted scripts/).

Note: the test suite was not run as part of this review — uv wasn't available in the review environment. Findings are from reading the code, git diff, and git check-ignore.

🤖 Generated with Claude Code

PLAN.md: pin down the contracts where the Frontend and Backend agents
would each have invented an incompatible answer — SSE frame shape,
error envelope, chat actions schema, priced-set union rule, trade
execution rules — and record the reasoning in a resolution log. Four
open questions for the project owner remain (Q1 blocks frontend work).

SKILL.md: rename the skill frontmatter cerebras-inference -> cerebras
so PLAN section 9's reference actually resolves.

Add the reviewer subagent and doc-review command that produced the
PLAN revision, so the workflow is reproducible from a fresh clone, plus
REVIEW.md recording the current review findings.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@kingryad kingryad closed this by deleting the head repository Aug 3, 2026
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