Skip to content

Slim agent infrastructure: navigation policy, rubric framework, RIG, cleanup#711

Open
johnml1135 wants to merge 4 commits intomainfrom
slim_agents
Open

Slim agent infrastructure: navigation policy, rubric framework, RIG, cleanup#711
johnml1135 wants to merge 4 commits intomainfrom
slim_agents

Conversation

@johnml1135
Copy link
Contributor

@johnml1135 johnml1135 commented Feb 25, 2026

Summary

Overhaul the AI agent infrastructure to be leaner, more structured, and research-informed.

Changes

Removed (~67 files)

  • Per-folder verbose AGENTS.md files throughout Src/ — replaced by centralized guidance
  • 10 Python hash/check/scaffold scripts (.github/*.py) — replaced by RIG
  • CI gate workflow for AGENTS docs
  • All Beads tooling, scripts, config, and assets

Added

  • Repository Intelligence Graph (RIG): Repository.Intelligence.Graph.json + generator script Build/Agent/Generate-RepositoryIntelligenceGraph.ps1 — deterministic project/build/test topology map
  • Navigation policy: .github/instructions/navigation.instructions.md — structural vs semantic task classification, Veto protocol for hidden dependencies, checklist-at-END placement (informed by CodeCompass arXiv:2602.20048)
  • Rubric framework (.github/rubrics/):
    • Base schema v2 with 4 equally-weighted categories (file_change, spec_alignment, integrity, runtime), 3 hard gates, 16 atomic criteria with evidence fields
    • 4 specializations: interop-boundary, installer-deployment, parser-grammar, ui-build-workflow
    • Design informed by Agentic Rubrics arXiv:2601.04171 with key improvements: atomic single-predicate criteria, mandatory evidence fields, outcome-based checks, focusCriteria boost semantics
  • Rubric verification skill: .github/skills/rubric-verify/SKILL.md — 5-step workflow (select rubric → identify evidence → score → gates → verdict)

Modified

  • AGENTS.md — simplified, added RIG/navigation/Serena references
  • .github/AGENTS.md — added navigation + rubric references
  • .github/AI_GOVERNANCE.md — added navigation.instructions.md to curated keep set
  • Skills (verify-test, review, jira-bugfix) — added rubric-verify cross-references
  • ~40 OpenSpec spec files — removed dead AGENTS.md links

Rubric quality improvements (v2)

Based on literature review of agentic evaluation research:

  • Atomicity: each criterion tests a single observable predicate (no compound "A and B" checks)
  • Evidence-grounded: every criterion has an evidence field specifying what artifact to cite
  • Balanced weights: all 4 categories at 0.25 (runtime was underweighted at 0.20)
  • Hard gates are mechanically verifiable: build exits 0, no new unsanitized boundary input, evidence present — not subjective judgments
  • focusCriteria semantics defined: 1.5x weight boost for specialization-relevant criteria
  • requiredCommands at base level: build.ps1 + test.ps1 always required
  • Renamed for clarity: deterministic_verification + environment_parityverification_reproducibility; marshalling_risk_reviewedmarshalling_verified (outcome, not process)
  • Added: test_coverage criterion, no_regression criterion

Testing

Agent infrastructure changes — no runtime code affected. Validated YAML structure and cross-references.


This change is Reviewable

@github-actions

This comment has been minimized.

@github-actions
Copy link

github-actions bot commented Feb 25, 2026

NUnit Tests

    1 files  ±0      1 suites  ±0   5m 53s ⏱️ -12s
4 407 tests ±0  4 320 ✅ ±0  87 💤 ±0  0 ❌ ±0 
4 416 runs  ±0  4 329 ✅ ±0  87 💤 ±0  0 ❌ ±0 

Results for commit 7eb2569. ± Comparison against base commit db65176.

♻️ This comment has been updated with latest results.

@johnml1135 johnml1135 marked this pull request as ready for review February 26, 2026 19:48
Copilot AI review requested due to automatic review settings February 26, 2026 19:48
Copy link

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Refactors the repo’s AI/agent infrastructure to a slimmer, centralized model by removing per-folder AGENTS sprawl and Beads tooling, and introducing a deterministic Repository Intelligence Graph (RIG), navigation policy, and a weighted rubric framework for verification.

Changes:

  • Removed numerous Src/**/AGENTS.md files and related Python automation/CI gate in favor of centralized guidance.
  • Added a rubric framework + rubric-verify skill for evidence-grounded scoring and hard-gate verdicts.
  • Added navigation policy and updated docs/tooling (README, VS Code tasks) to point to RIG regeneration and the new processes.

Reviewed changes

Copilot reviewed 121 out of 174 changed files in this pull request and generated 2 comments.

Show a summary per file
File Description
Src/xWorks/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/XCore/xCoreTests/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/XCore/xCoreInterfaces/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/XCore/SilSidePane/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/XCore/FlexUIAdapter/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/XCore/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/Utilities/XMLUtils/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/Utilities/SfmToXml/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/Utilities/SfmStats/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/Utilities/Reporting/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/Utilities/MessageBoxExLib/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/Utilities/FixFwData/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/Utilities/AGENTS.md Removes parent-folder AGENTS overview (centralization).
Src/UnicodeCharEditor/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/ProjectUnpacker/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/ParatextImport/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/ManagedVwWindow/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/ManagedVwDrawRootBuffered/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/LexText/LexTextDll/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/LexText/FlexPathwayPlugin/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/LexText/AGENTS.md Removes parent-folder AGENTS overview (centralization).
Src/Kernel/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/InstallValidator/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/Generic/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/GenerateHCConfig/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/FwResources/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/FwParatextLexiconPlugin/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/FwCoreDlgs/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/FdoUi/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/FXT/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/DocConvert/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/DebugProcs/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/DbExtend/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/Common/ViewsInterfaces/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/Common/UIAdapterInterfaces/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/Common/ScriptureUtils/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/Common/RootSite/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/Common/FwUtils/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/Common/Controls/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/Common/AGENTS.md Removes parent-folder AGENTS overview (centralization).
Src/Cellar/AGENTS.md Removes per-folder AGENTS documentation (centralization).
Src/AGENTS.md Adds minimal, centralized guidance for work under Src/.
ReadMe.md Updates agent-docs section to reference the minimal AGENTS model + RIG regeneration.
FLExInstaller/wix6/AGENTS.md Removes WiX6 subfolder AGENTS guidance (centralization).
FLExInstaller/AGENTS.md Replaces verbose installer AGENTS with a minimal “defaults/constraints” guide.
Docs/agent-docs-refresh.md Rewrites refresh workflow to match the minimal AGENTS + RIG model.
.vscode/tasks.json Removes Beads/Jira tasks and adds an “Agent: Generate RIG” task.
.github/workflows/copilot-docs-detect.yml Removes the agent-docs detection workflow.
.github/skills/verify-test/SKILL.md Cross-references rubric-verify as an option for execution-free scoring.
.github/skills/rubric-verify/SKILL.md Adds rubric verification skill workflow with evidence and hard gates.
.github/skills/review/SKILL.md Cross-references rubric-verify for scoring/hard-gate assessment.
.github/skills/openspec-beads-import/SKILL.md Removes the OpenSpec→Beads import skill.
.github/skills/jira-to-beads/scripts/export_jira_assigned.py Removes Jira export script as part of Beads/tooling removal.
.github/skills/jira-to-beads/SKILL.md Removes Jira→Beads skill.
.github/skills/execute-implement/SKILL.md Updates notes, removing Beads references and pointing to Jira/PR updates.
.github/skills/beads/resources/WORKTREES.md Removes Beads resource documentation.
.github/skills/beads/resources/STATIC_DATA.md Removes Beads resource documentation.
.github/skills/beads/resources/RESUMABILITY.md Removes Beads resource documentation.
.github/skills/beads/resources/ISSUE_CREATION.md Removes Beads resource documentation.
.github/skills/beads/resources/CHEMISTRY_PATTERNS.md Removes Beads resource documentation.
.github/skills/beads/resources/ASYNC_GATES.md Removes Beads resource documentation.
.github/skills/beads/resources/AGENTS.md Removes Beads resource documentation.
.github/skills/beads/adr/0001-bd-prime-as-source-of-truth.md Removes Beads ADR.
.github/skills/beads/SKILL.md Removes Beads skill.
.github/skills/beads/README.md Removes Beads skill README.
.github/skills/beads/CLAUDE.md Removes Beads skill maintenance doc.
.github/skills/atlassian-readonly-skills/SKILL.md Removes references to the deleted jira-to-beads helper.
.github/rubrics/ui-build-workflow.rubric.yaml Adds specialized rubric for UI/build workflow changes.
.github/rubrics/parser-grammar.rubric.yaml Adds specialized rubric for parser/grammar changes.
.github/rubrics/interop-boundary.rubric.yaml Adds specialized rubric for native/managed/COM boundary changes.
.github/rubrics/installer-deployment.rubric.yaml Adds specialized rubric for installer/deployment changes.
.github/rubrics/fieldworks-rubric.base.yaml Adds the base rubric schema v2 with weighted categories and hard gates.
.github/migrate_copilot_format.py Removes legacy AGENTS migration helper.
.github/instructions/terminal.instructions.md Removes Beads CLI auto-approval guidance.
.github/instructions/navigation.instructions.md Adds the navigation policy for structural vs semantic tasks + veto protocol.
.github/copilot_tree_hash.py Removes legacy tree-hash helper used by the old AGENTS pipeline.
.github/copilot_doc_utils.py Removes legacy markdown helper utilities for the old AGENTS pipeline.
.github/copilot_change_utils.py Removes legacy change classification utilities for the old AGENTS pipeline.
.github/copilot_cache.py Removes legacy cache helper used by the old AGENTS pipeline.
.github/copilot_apply_updates.py Removes legacy auto-section application script for old per-folder AGENTS.
.github/AI_GOVERNANCE.md Adds navigation.instructions.md to the curated keep set.
.gitattributes Removes Beads merge driver config (consistent with Beads removal).

- Remove the extra agents and agent CI work
- Add a RIG
- Remove beads
- Add navigation policy, rubric framework, and verification skill
- Add .github/instructions/navigation.instructions.md: structural-vs-semantic
  task classification, Veto protocol for hidden dependencies, checklist-at-END
  placement guidance (informed by CodeCompass arXiv:2602.20048)

- Add .github/rubrics/ with base schema v2 + 4 specializations:
  - fieldworks-rubric.base.yaml: 4-category weighted scoring (file_change,
    spec_alignment, integrity, runtime at 0.25 each), 3 hard gates, 16 criteria
    with evidence fields, focusCriteria boost semantics, requiredCommands
  - interop-boundary.rubric.yaml: native/managed/COM boundary focus
  - installer-deployment.rubric.yaml: WiX installer with snapshot evidence
  - parser-grammar.rubric.yaml: parsing/morphology rule traceability
  - ui-build-workflow.rubric.yaml: UI localization and build workflow
  Design informed by arXiv:2601.04171 (Agentic Rubrics) with improvements:
  atomic single-predicate criteria, evidence fields on every criterion,
  outcome-based checks (not process checks), balanced category weights

- Add .github/skills/rubric-verify/SKILL.md: 5-step workflow (select rubric,
  identify evidence, score criteria, evaluate hard gates, report verdict)

- Wire references into AGENTS.md, .github/AGENTS.md, AI_GOVERNANCE.md,
  verify-test, review, and jira-bugfix skills
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.

2 participants