refactor(frontend-oracle-design): graph-loaded reference nodes with shared common file and review-point links - #5
Merged
Conversation
…nodes with a shared common file
- Add references/reference-graph.json: every reference is a node with a load
condition (when) and co-load dependency edges (requires); SKILL.md loading
section now follows the graph so only the nodes a phase needs are read.
- Add references/common.md as the single canonical owner of the authority
priority order, recognized policy sources, jurisdiction rules, the six-way
feedback routing table, and shared prohibitions; changeability,
frontend-implementation, subagent-review, visual-design, and the delivery
nodes now point at it instead of duplicating those blocks.
- Split oracle-card.md into card/{policy-sources,risk-grill,card-format,
confirmation-lock}.md, implementation-loop.md into delivery/{ledger,red,
implementation-decision,green-review}.md, and type-constraints.md into
types/{state-ladder,authoring,api-surface,review-criteria}.md so each
phase loads only its slice.
- Keep bva.md byte-identical (shared with the test package) and keep
single-topic references at their existing paths.
- Update contract tests to read node concatenations, validate the graph
(unique ids, resolvable edges, complete file coverage), and assert the
common-file dedup; release 0.18.0.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RRu1pxas722y7uzbER5CL3
…le links - Add repeatable --review-point to oracle-run.mjs review-packet: each point is validated (exists, regular file, non-empty, no duplicates) and recorded in the packet as path + SHA-256 only — the reviewer follows the link and reads the file itself, and the digest pins which criteria revision was used. - Document the review-point routing table in subagent-review.md (diff condition → criteria file link) and forbid pasting criteria text into the reviewer prompt; SKILL.md review loading bullet points at the flag. - Declare the same condition → node routing machine-readably as reviewPoints in reference-graph.json. - Cover with a functional oracle-run test (link-only packet field, invalid/ empty/duplicate points rejected) and a contract test (docs, flag, graph routing resolves); release 0.18.1. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RRu1pxas722y7uzbER5CL3
…erence graph as an exclusive lane - Add references/lanes/low-fast-path.md owning the lane contract: entry conditions, procedure (repo verifications still run), what is skipped (card/lock/ledger/review), and explicit disqualification-escalation rules into the oracle lane. - Declare a machine-readable lanes section in reference-graph.json: the Low lane loads only its node (exclusive) with an escalation rule; the oracle lane enters at common. - SKILL.md now routes risk=Low to the lane node in both the entry rule and the loading graph, deduplicating the previous double description; card/risk-grill.md and delivery/ledger.md link the lane instead of restating it. - Contract test covers the lane file sections, graph lane resolution, and routing links; release 0.18.2. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RRu1pxas722y7uzbER5CL3
…anonical owners - SKILL.md feedback-routing section keeps the six classification names but drops the arrow table duplicated verbatim in common.md. - card/risk-grill.md UI design gate keeps the load trigger and lock rule but drops the bullets restating visual-design.md's proposal/two-pass/--source rules, which that file owns. - subagent-review.md checklist drops the type/state items now delivered as review-point links (types/review-criteria.md, frontend-implementation.md) and replaces them with a single do-not-repeat pointer item. - Update the two affected test pins; release 0.18.3. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RRu1pxas722y7uzbER5CL3
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 9a7e4de4f3
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| } | ||
| // 링크만 전달한다 — reviewer가 경로의 파일을 직접 전부 읽고, digest로 어떤 | ||
| // revision의 기준을 읽었는지 고정한다. 본문을 packet에 복제하지 않는다. | ||
| reviewPoints.push({ path: point, sha256: sha256(content) }) |
There was a problem hiding this comment.
When --review-point is supplied as a relative path and the reviewer reads the packet from a different working directory, validation hashes pointPath resolved against the generator's cwd but the packet stores the original relative point. The reviewer can therefore fail to open the criterion or read a different file, so the recorded digest no longer fixes the linked review input. Store the resolved/canonical path, or make it relative to a stable packet or scan root.
Useful? React with 👍 / 👎.
…the split's line overhead - Serialize oracle-workflow.graph.json edges one per line (content-identical JSON, verified by parse equality; graph-verify still passes): 602 -> 317 lines. - Slim reference-graph.json: drop the decorative phase field and shorten node when strings to compact phrases — the full load-condition sentences are owned by SKILL.md's loading section, so the duplication goes away too. - Normalize all skill markdown/json through the repo prettier config. - Net: references + SKILL total is 3,971 lines vs 3,931 before the split (+40), while per-stage loaded context stays reduced; release 0.18.4. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RRu1pxas722y7uzbER5CL3
lodado
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What changed
oracle-card.md→card/{policy-sources,risk-grill,card-format,confirmation-lock}.md,implementation-loop.md→delivery/{ledger,red,implementation-decision,green-review}.md,type-constraints.md→types/{state-ladder,authoring,api-surface,review-criteria}.md.references/reference-graph.json: every reference is a node with a load condition (when) and co-load dependency edges (requires);SKILL.md's loading section follows the graph so each phase reads only its nodes.references/common.mdas the single canonical owner of the authority priority order, recognized policy sources, jurisdiction rules, the six-way feedback routing table, and shared prohibitions; other references now point at it instead of duplicating those blocks.--review-pointonoracle-run.mjs review-packetrecords path + SHA-256 only (no pasted text), with the condition → criteria-file routing table insubagent-review.mdand machine-readablereviewPointsin the graph.references/lanes/low-fast-path.md) with entry conditions and disqualification-escalation rules.bva.mdbyte-identical with thetestpackage. Released 0.18.4.Why
How tested
pnpm test)pnpm build)pnpm typecheck)pnpm lint)Test details
packages/frontend-oracle-design:node --test skills/scripts/*.test.mjs— 121 pass, 0 fail (contract tests updated to read node concatenations; new tests validate the graph — unique ids, resolvable edges, complete file coverage — the common-file dedup, review-point link recording, and the low-fast-path lane).packages/test:node --test scripts/*.test.mjs— 3 pass, 0 fail (bva.mdbyte-equality with the oracle package preserved).oracle-run.mjs review-packet --review-pointfunctional test: link-only packet field (path + sha256, no content), missing/empty/duplicate points rejected withREVIEW_POINT_INVALID.oracle-workflow.graph.jsonre-serialization verified content-identical by JSON parse equality; bundledgraph-verifystill returnsGRAPH_VALID.Risk
Affected areas
packages/sdui-template)packages/sdui-template-component)apps/docs)apps/nextAuthOauthLoginExample)packages/frontend-oracle-design(skill references, contract tests,oracle-run.mjs),.claude-plugin/marketplace.jsonRollback plan
git revert <merge-commit-hash>Checklist
🤖 Generated with Claude Code
https://claude.ai/code/session_01RRu1pxas722y7uzbER5CL3