Repository navigation
fix(instructions): repair four R3 contract defects across all plugin targets (ai-generated, includes-ai-code) - #368
Open
ArteomBalanuta wants to merge 4 commits into
Conversation
…gins Closes griddynamics#357, Closes griddynamics#355, Closes griddynamics#354, Closes griddynamics#356 One regeneration pass propagates four independent R3 source repairs to all 14 plugin targets (standard + lightweight): - griddynamics#357: normalize the state-path token to coding-agents-prompting-flow-state.md and the step-7 producer/consumer identifier (Prompt Set, not Candidate Prompt Set) — resume and final validation handoff were broken in every distribution. - griddynamics#355: replace the unconditional DeepWiki+Context7 mandate with a capability- aware source contract (prefer when available, explicit fallbacks otherwise); research README updated to match. - griddynamics#354: remove nested subagent dispatch from the research skill, research-flow phase 3, and the research README — the researcher executes the approved prompt directly; parallel-slice needs are proposed back to the top-level orchestrator. - griddynamics#356: define the small in-place modernization reduced route as an explicit, auditable exception to the phase-prerequisite contract (Phase 2 without Phase 1 only when no reference-library analysis is needed; Phase 6 without Phase 5 only under a bounded single-project final-review note). R2 (KTLO) carries equivalent text and is intentionally untouched — an R2 backport is a separate release-policy decision.
ArteomBalanuta
requested review from
isolomatov-gd and
omaiesh
as code owners
October 15, 2026 15:30
github-actions
Bot
added
bug
Contributor
Rosetta Triage ReviewSummary: Fixes four R3 instruction defects in prompt-authoring, research, and modernization flows. The PR updates all 14 generated plugin targets. Classification: bug · instructions · needs more work Findings:
Caveats:
Automated triage by Rosetta agent |
…alidate Prompts) The prompt-quality auditor flagged a High conflict: the unconditional 'execute directly, never spawn subagents' rule broke the top-level team-manager caller, which activates research directly and must keep dispatching the prepared prompt to a researcher subagent. Rule is now executor-scoped: spawned researcher executes locally and proposes parallel slices to its parent; top-level agent dispatches under orchestration. Nested dispatch stays prohibited. Regenerated across all 14 plugin targets.
ArteomBalanuta
commented
Oct 15, 2026
ArteomBalanuta
left a comment
ArteomBalanuta
left a comment
Contributor
Author
There was a problem hiding this comment.
Review notes (left as a comment, not approve/request-changes):
- #357 (prompting-flow state path and the
Prompt Setidentifier): looks correct. Every other reference already uses.md, and nothing else in R3 saysCandidate Prompt Set. - Plugin regeneration looks consistent. All 70 plugin files carry only lines from the 5 source changes, and the totals match exactly (154 added = 14 × 11, 140 removed = 14 × 10).
- Not mergeable as-is: the #356 modernization exception still conflicts with the phase files, and the #354 research change blocks top-level callers (see inline comments). The
validatecheck is failing on the second point. - Leaving R2 unchanged is fine under the KTLO policy.
ArteomBalanuta
marked this pull request as draft
October 15, 2026 15:48
…rch contracts - griddynamics#356 follow-through: the reduced route now defines its substitutes IN the phase files the agent loads — modernization-flow-analysis.md and modernization-flow-mapping.md carry explicit reduced-route exceptions on their prerequisites and MUST re-read lines; the sequential-execution note is qualified; the state-file template gains a Reduced Route field so the exception is auditable; Phase 7 is explicitly skipped on the reduced route (Phase 6 single-project note is its substitute); 'final-review' wording corrected to 'single-project mapping scope'. - research SKILL.md: removed the capability-aware source line that duplicated enforcement rule 2; role wording aligned ('execute that prompt — never ad-hoc research'). - research README: self-contradictory meta-prompting line rewritten to the caller-scoped contract; stale line citation corrected (research-flow.md:50). Regenerated across all 14 plugin targets; 797 generator tests pass.
ArteomBalanuta
marked this pull request as ready for review
October 22, 2026 19:39
…n flow - Phase 8 (implement) prerequisite now carries the reduced-route exception: Phase 7 is skipped there, so Phase 6 + the single-project scope note stand in for the cross-project review. - "DO NOT SKIP PHASES" / "Do not skip phases!" in Critical Requirements now name the reduced route as the only exception. - The single-project scope note has a defined home: the orchestrator writes it in the state file's Reduced Route section before Phase 6; Phase 6 prerequisites and its re-read requirement point there. Regenerated across all 14 plugin targets; 797 generator tests pass. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Contributor
Rosetta Triage ReviewSummary: Repairs R3 instruction defects and regenerates plugin outputs. The Phase 2 and Phase 6 prerequisites now include the reduced route. Classification: bug · instructions Findings:
Suggestions:
Clarifications:
Caveats:
Automated triage by Rosetta agent |
Contributor
📋 Prompt Quality Validation Report✅ Validation PassedThe full markdown report and raw JSON output are available in the workflow artifacts for 5 days. No issues were found in the changed prompt files. |
This branch has not been deployed
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
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.
Summary
Jira: N/A (GitHub issues #357, #355, #354, #356)
Closes #357.
Closes #355.
Closes #354.
Closes #356.
Four confirmed defects in the canonical R3 instruction sources, all shipping inside every generated plugin distribution. Each was verified against current source before repair:
.md(line 29) while every phase updates the.mdform, and step 7 consumedCandidate Prompt Setalthough step 5 producesPrompt Set. Both producer/consumer identifiers normalized; the seven-step handoff now simulates end to end.What changed
instructions/r3/core/workflows/coding-agents-prompting-flow.md,research-flow.md,modernization-flow.md,skills/research/SKILL.md,skills/research/README.md).rosettify-pluginsstandard + lightweight passes,--deterministic-hooks false, matchingscripts/pre_commit.py's pinned invocation).Test plan
Candidate Prompt Set,MUST use DeepWiki and Context7,Spawn parallel subagents, or state-path-without-.mdin R3 sources and all plugin outputs; fixed text present in all 14 plugin targets.npm test(rosettify-plugins) — 797 passed / 52 files.Prompt Set→ step 6 consumes it → step 7 consumes it and producesFinal Prompt Set; state path consistent across all seven phase updates and the checklist.