Cutout skills from flows

.github/workflows/validate-prompts.yml

@@ -41,6 +41,7 @@ jobs:
       TMP_DIR: .tmp
       ALL_RESULTS_FILE: .tmp/all-results.json
       PR_COMMENT_FILE: .tmp/pr-comment.md
+      PR_DETAILS_FILE: .tmp/pr-details.md
       DIFF_FILE: .tmp/instructions.diff
       MODEL: opus
 
@@ -172,31 +173,45 @@ jobs:
             fi
           done
 
-          echo -e "\n---\n" >> $PR_COMMENT_FILE
+          # PR comment intentionally stops at the summary table — GitHub PR comments are capped at 65,536 chars
+          # and per-file detail blocks easily blow past that on multi-file PRs. Full per-file findings go to the
+          # workflow's $GITHUB_STEP_SUMMARY (no size limit) via $PR_DETAILS_FILE; the PR comment links to it.
+          RUN_URL="${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+          cat >> $PR_COMMENT_FILE <<EOF
+
+          ---
+
+          :clipboard: **Full per-file findings (Problem / Reason / Solution + Gates Comparison) → [Workflow run Summary]($RUN_URL)** (PR comments are capped at 65,536 chars; details live on the Actions run).
+          EOF
+
+          # Per-file detail blocks: write to $PR_DETAILS_FILE (consumed by the Summary step → $GITHUB_STEP_SUMMARY).
+          cat > $PR_DETAILS_FILE <<EOF
+          ## :clipboard: Per-File Findings
+
+          EOF
 
-          # Process each file with detailed issues
           jq -c '.[]' $ALL_RESULTS_FILE | while read -r result; do
             FILE=$(echo "$result" | jq -r '.file')
 
             # Show file name first
-            echo -e "### :page_facing_up: \`$FILE\`\n" >> $PR_COMMENT_FILE
+            echo -e "### :page_facing_up: \`$FILE\`\n" >> $PR_DETAILS_FILE
 
             if echo "$result" | jq -e '.error' > /dev/null 2>&1; then
-              echo "**Error:** $(echo "$result" | jq -r '.error')" >> $PR_COMMENT_FILE
-              echo -e "\n---\n" >> $PR_COMMENT_FILE
+              echo "**Error:** $(echo "$result" | jq -r '.error')" >> $PR_DETAILS_FILE
+              echo -e "\n---\n" >> $PR_DETAILS_FILE
               continue
             fi
 
             # Show issues
             ISSUES=$(echo "$result" | jq '[.issues | sort_by(.severity) | reverse[]]')
 
             if [ "$(echo "$ISSUES" | jq 'length')" -eq 0 ]; then
-              echo "#### :white_check_mark: No Issues Found" >> $PR_COMMENT_FILE
-              echo "" >> $PR_COMMENT_FILE
+              echo "#### :white_check_mark: No Issues Found" >> $PR_DETAILS_FILE
+              echo "" >> $PR_DETAILS_FILE
             else
-              echo "#### :warning: Issues Found" >> $PR_COMMENT_FILE
-              echo "" >> $PR_COMMENT_FILE
-              cat >> $PR_COMMENT_FILE <<EOF
+              echo "#### :warning: Issues Found" >> $PR_DETAILS_FILE
+              echo "" >> $PR_DETAILS_FILE
+              cat >> $PR_DETAILS_FILE <<EOF
           | Severity | Gate | Details |
           |----------|------|---------|
           EOF
@@ -209,17 +224,17 @@ jobs:
                  else ":black_circle: Unknown" end),
                 .gate, .problem, .reason, .solution] | @tsv' | \
               while IFS=$'\t' read -r sev gate prob reason sol; do
-                echo "| $sev | $gate | **Problem:**<br> $prob<br>**Reason:**<br> $reason<br>**Solution:**<br> $sol |" >> $PR_COMMENT_FILE
+                echo "| $sev | $gate | **Problem:**<br> $prob<br>**Reason:**<br> $reason<br>**Solution:**<br> $sol |" >> $PR_DETAILS_FILE
               done
-              echo "" >> $PR_COMMENT_FILE
+              echo "" >> $PR_DETAILS_FILE
             fi
 
             # Show gates comparison with scores
             echo "$result" | jq -e '.gates' > /dev/null 2>&1 && {
-              echo "#### :bar_chart: Gates Comparison" >> $PR_COMMENT_FILE
-              echo "" >> $PR_COMMENT_FILE
-              echo "| Gate | Score | Comparison |" >> $PR_COMMENT_FILE
-              echo "|------|-------|------------|" >> $PR_COMMENT_FILE
+              echo "#### :bar_chart: Gates Comparison" >> $PR_DETAILS_FILE
+              echo "" >> $PR_DETAILS_FILE
+              echo "| Gate | Score | Comparison |" >> $PR_DETAILS_FILE
+              echo "|------|-------|------------|" >> $PR_DETAILS_FILE
 
               echo "$result" | jq -r '.gates | to_entries | map(select(.value.comparison != 3)) | .[] |
                 [.key,
@@ -230,13 +245,13 @@ jobs:
                   elif .value.comparison == 1 then ":x: Much worse"
                   else ":arrow_right: No change" end)] | @tsv' | \
               while IFS=$'\t' read -r gate score change; do
-                echo "| $gate | $score | $change |" >> $PR_COMMENT_FILE
+                echo "| $gate | $score | $change |" >> $PR_DETAILS_FILE
               done
 
-              echo "" >> $PR_COMMENT_FILE
+              echo "" >> $PR_DETAILS_FILE
             }
 
-            echo -e "\n---\n" >> $PR_COMMENT_FILE
+            echo -e "\n---\n" >> $PR_DETAILS_FILE
           done
 
       - name: Post PR comment
@@ -260,4 +275,10 @@ jobs:
 
           - **Files validated:** ${{ steps.changed-files.outputs.count }}
           - **Status:** $( [ "${{ steps.validate.outputs.has_high_severity }}" = "true" ] && echo ":x: Failed" || echo ":white_check_mark: Passed" )
+
           EOF
+
+          # Append the compact PR comment (summary table) AND the full per-file detail block to the run summary.
+          # The PR comment posts only the compact part; the full detail lives here without the 65,536-char cap.
+          [ -f "$PR_COMMENT_FILE" ] && cat "$PR_COMMENT_FILE" >> $GITHUB_STEP_SUMMARY
+          [ -f "$PR_DETAILS_FILE" ] && cat "$PR_DETAILS_FILE" >> $GITHUB_STEP_SUMMARY

CHANGELOG.md

@@ -1,5 +1,76 @@
 # Changelog
 
+## R3 Migration Guide
+
+**Release:** R3 (2026-06-01)
+
+This guide covers breaking changes between R2 and R3 for prompt and workflow authors, plugin packagers, and anyone who ACQUIRE's Rosetta instructions by name. No compatibility shims are provided; cutover is hard at the R3 release boundary.
+
+---
+
+### 1. `plan-manager` skill renamed to `operation-manager`
+
+- **What changed.** The R2 skill `plan-manager` is removed in R3 and replaced by `operation-manager` at `instructions/r3/core/skills/operation-manager/`. The skill scope is unchanged (plan create / next / update_status / query / show_status / upsert), but the canonical name, command alias (`OPERATION_MANAGER`), and bootstrap references all use the new name. The schema asset was renamed `pm-schema.md` → `om-schema.md`.
+- **Who is affected.** Prompt authors, workflow authors, and any rule/skill that ACQUIRE's `plan-manager` by name or references the `PLAN_MANAGER` alias.
+- **Required action.** Replace name and alias across authored content:
+  ```text
+  ACQUIRE plan-manager FROM KB        ->  ACQUIRE operation-manager FROM KB
+  PLAN_MANAGER                        ->  OPERATION_MANAGER
+  plan-manager next <plan_file>       ->  operation-manager next <plan_file>
+  plan-manager upsert ...             ->  operation-manager upsert-with-template ...
+  pm-schema.md                        ->  om-schema.md
+  ```
+  Recommended grep before merging:
+  ```bash
+  grep -rn -e 'plan-manager' -e 'PLAN_MANAGER' -e 'pm-schema.md' .
+  ```
+- **Rollout / cutover note.** Hard cutover at R3. No alias, no deprecation window, no shim. R2 consumers stay on the `plan-manager` skill until they upgrade. Any residual `plan-manager` reference in an R3 authored prompt is a bug and will fail to resolve through Rosetta MCP `query_instructions`.
+
+---
+
+### 2. `bootstrap-hitl-questioning.md` removed
+
+- **What changed.** The bootstrap rule `bootstrap-hitl-questioning.md` (present in R2 at `instructions/r2/core/rules/`) is removed in R3. HITL enforcement is now consolidated into the on-demand `hitl` skill at `instructions/r3/core/skills/hitl/`, and the entry point is `bootstrap-guardrails.md`, which references the skill. This is part of the bootstrap size reduction shipped in R3.
+- **Who is affected.** Authors who linked to `bootstrap-hitl-questioning.md` by path or filename, plugin packagers who copied the rule file into IDE bundles, and any rule/workflow that ACQUIRE's HITL questioning content by the old filename.
+- **Required action.** Replace direct references to the deleted rule with a skill invocation:
+  ```text
+  ACQUIRE bootstrap-hitl-questioning FROM KB
+      ->  USE SKILL `hitl`
+
+  instructions/.../rules/bootstrap-hitl-questioning.md
+      ->  instructions/r3/core/skills/hitl/SKILL.md  (load via the skill, not by path)
+  ```
+  The guardrails entry point already wires this in — authors typically only need to remove the explicit reference and trust `bootstrap-guardrails.md` to route to the skill on demand.
+- **Known coverage gaps** to be aware of when migrating: six items from the R2 `bootstrap-hitl-questioning.md` were not fully ported into the R3 `hitl` skill — see [docs/TODO.md](docs/TODO.md) "hitl skill — R2 coverage gaps" for the enumerated list. None of these block R3 release, but downstream consumers relying on R2-specific HITL phrasing (graduated MEDIUM/HIGH/CRITICAL escalation, cognitive-load limits, mismatch root-cause memory update) should review before upgrading.
+- **Rollout / cutover note.** Hard removal in R3. No file at the old path. Plugin bundles that still ship the R2 file should drop it on their next sync; the `hitl` skill folder is included in every R3 plugin under `skills/hitl/`.
+
+---
+
+### 3. New skill families added in R3
+
+- **What changed.** R3 introduces three new skill families under `instructions/r3/core/skills/`. Plugin packagers must include these folders in downstream bundles or the corresponding workflows will fail to resolve at runtime.
+  - **QA workflow family.** Skills supporting the QA, AQA, and TestGen workflows: `qa-*`, `aqa-*`, `automation-*`, `api-test-spec-authoring`, `testrail-*`, `mcp-jira-data-collection`, `mcp-confluence-data-collection`, `mcp-testrail-data-collection`, `swagger-contracts-analysis`, `confluence-source-harvesting`, `aqa-requirements-elicitation`, `gap-and-contradiction-analysis`.
+  - **Utility skills.** General process and authoring skills new in R3: `sequential-workflow-execution`, `requirements-synthesis`, `user-approved-code-changes`, `repository-implementation-standards`, `load-context-instructions`, `load-workflow`.
+  - **GitNexus tools.** Code-graph integration skills: `gitnexus-setup`, `gitnexus-cli`, `gitnexus-tools`.
+- **Who is affected.** Plugin packagers (`core-claude`, `core-cursor`, `core-cursor-standalone`, `core-copilot`, `core-copilot-standalone`, `core-codex`) and any downstream distribution that copies skill folders selectively rather than mirroring `instructions/r3/core/skills/` wholesale.
+- **Required action.** Re-run the plugin sync so the new skill folders are picked up. After sync, verify each plugin's `skills/` directory contains the three new families above. For selective packagers, add the family prefixes (`qa-*`, `aqa-*`, `automation-*`, `mcp-*-data-collection`, `testrail-*`, `gitnexus-*`) to the include list.
+- **Rollout / cutover note.** Additive only — no removals in this item. Workflows that depend on these skills (QA, AQA, TestGen, GitNexus-enabled flows) will not function in a plugin that lacks the folders. Mirror the full `skills/` tree if in doubt.
+
+---
+
+### How to apply
+
+1. **Re-run plugin sync.** Regenerate every plugin bundle so the renamed `operation-manager` skill, the removed `bootstrap-hitl-questioning.md`, and the new skill families are picked up consistently. Note that `plugin_generator.py` `DEFAULT_RELEASE` is now `r3` (was `r2`); existing CI / pre-commit invocations no longer need to pass `--release r3` explicitly.
+2. **Grep for stale names.** From the repo root:
+   ```bash
+   grep -rn -e 'plan-manager' -e 'PLAN_MANAGER' -e 'bootstrap-hitl-questioning' -e 'pm-schema.md' .
+   ```
+   Any hit in R3 authored content or in a plugin bundle is a defect — fix before release.
+3. **ACQUIRE re-check.** For each authored prompt or workflow that previously ran `ACQUIRE plan-manager` or `ACQUIRE bootstrap-hitl-questioning`, re-resolve against the R3 KB and confirm the new name (`operation-manager`) or the new path (`hitl` skill) returns a document.
+4. **Smoke test workflows.** Run one workflow per affected family (operation-manager driven plan, a HITL gated step, one QA workflow, one GitNexus enabled session) to confirm end-to-end resolution before announcing the cutover.
+
+---
+
 ## R2
 
 ### Overview