Skip to content

V0.11.0/chore and git remote pr - #44

Merged
gimlichael merged 48 commits into
mainfrom
v0.11.0/chore-and-git-remote-pr
Sep 24, 2026
Merged

gimlichael merged 48 commits into
mainfrom
v0.11.0/chore-and-git-remote-pr

Conversation

@gimlichael

@gimlichael gimlichael commented Sep 23, 2026 •

Copy link
Copy Markdown
Member

This pull request introduces the deterministic git-remote-pr skill for creating and refreshing GitHub pull requests from committed changesets, establishes repository standards for scratch isolation, prose formatting, and PR routing, refactors agent-smith around an operating model framework, and adds visual identification assets across the skill catalog.

New git-remote-pr skill:

  • Creates or updates one PR from the full committed branch comparison using read-only evidence collection, theme coverage checks, and body structure validation before any remote write.
  • Requires explicit approval before push or metadata writes; same-request yolo/auto proceeds after showing the same preview.
  • Blocks on dirty worktrees, missing or mismatched upstream tracking, and diverged remote heads; verifies title, body, draft state, assignee, and file inventory after execution.
  • Ships prepare, plan, and execute helpers with regression coverage for CREATE/UPDATE distinction, approval integrity, and malformed-body rejection.

Repository standards:

  • Adds PR Skill Routing, Working-tree Scratch Isolation, and Markdown Prose Formatting guidance to repository and skill documentation.
  • Aligns commit, release, and squash-summary skills so temporary artifacts stay outside the working tree and prose stays on continuous physical lines.
  • Documents the new workflow, standards, and v0.11.0 changelog entries for reviewers and skill authors.

Agent-smith operating model:

  • Restructures the skill around a four-level decision framework with local repository policy replacing repository-family conventions.
  • Adds durable software and agentic automation reference guides; refreshes core principles, delivery, API, and skill-authoring guidance.
  • Updates evaluation definitions to match the operating model and precedence rules.

Documentation consistency:

  • Applies continuous-line prose formatting and description refinements across the skill portfolio, scaffold templates, and NuGet references.
  • Clarifies trigger-oriented skill descriptions and hard-wrap expectations without changing scaffold behavior.

Validation coverage:

  • Extends template validation for agent-smith policy checks, markdown formatting assertions, and git-remote-pr integration.
  • Adds agent-smith validation and historical regression isolation suites under temporary workspaces.

Visual skill identification:

  • Adds hero images for eight skills to support catalog rendering and visual identification.

aicia-bot and others added 15 commits September 15, 2026 23:24
Enforce a consistent working-tree scratch isolation principle across git-visual-commits, git-remote-release, and git-visual-squash-summary skills. These changes add explicit guidance that agents should never create or leave temporary files or directories in the repository, instead using operating-system temporary directories or session storage outside the repository. This principle prevents repository pollution and maintains clear separation between repository subjects and workflow artifacts.
Add evaluation case #29 to verify that git-visual-commits respects the working-tree scratch isolation principle. This eval tests that the skill performs plan-only review without creating repo-relative scratch files, instead using terminal output or verified operating-system temp storage when persistence is necessary.
Add validation assertions to ensure that working-tree scratch isolation guidance is present in git-visual-commits, git-remote-release, and git-visual-squash-summary skills. This includes assertions for the new Working-tree Scratch Isolation section in git-visual-commits, updated guidance in git-remote-release removing repository-local storage recommendations, and new non-mutating guidance in git-visual-squash-summary. Also validates that the corresponding eval case exists in git-visual-commits evals.
Clarify pre-write requirements for scratch file handling: verify temporary paths before writing, never use project-local paths even with cleanup planned, check resolved physical location including symlink targets, provide specific examples of forbidden patterns like git_diff_output.txt, and add corresponding eval assertions.
Add assertions to validate that the git-visual-commits skill documentation explicitly covers scratch file isolation requirements: pre-write verification, symlink/junction target checks, forbidden patterns, and terminal-only fallback behavior for truncated output.
Repository policies should be explicit and unambiguous. This change clarifies the Markdown prose formatting standard in AGENTS.md, establishing that paragraphs and list items should remain on continuous physical lines without arbitrary hard wrapping, relying on editor soft wrapping for visual presentation instead.
README.md now includes guidance for skill authors about the Markdown prose formatting standard. This communicates the expected formatting approach (continuous physical lines, soft wrapping, no arbitrary hard breaks) to users who contribute to or maintain skills.
Restructure skill documentation across 9 different skills to incorporate the Markdown prose formatting standard. This includes updates to SKILL.md files, skill references, evaluation definitions, and scaffold asset templates. Each prose paragraph and list item is kept on one physical line, eliminating arbitrary hard wraps in favor of editor soft wrapping. This ensures consistency with the repository standards and improves maintainability across all skill documentation.
The validation suite needs to verify that the new Markdown prose formatting standard is being followed correctly. This update changes the test assertions in validate-skill-templates.ps1 to reflect the new requirement that prose paragraphs and bullet items remain on continuous physical lines without arbitrary hard wrapping.
Release v0.11.0 introduces git-remote-pr, a deterministic Git and GitHub CLI
skill for managing pull requests from complete committed changesets. The
release establishes three core repository standards: working-tree scratch
isolation (keeping temporary artifacts outside the repository), markdown
prose formatting conventions (continuous physical lines without arbitrary
hard wrapping), and PR skill routing guidance in AGENTS.md.

The commit includes:
- git-remote-pr skill with prepare-pr.ps1, make-plan.ps1, execute-pr.ps1,
  and test harness
- AGENTS.md updates: PR Skill Routing, Working-tree Scratch Isolation, and
  Markdown Prose Formatting sections
- README.md updates: git-remote-pr catalog entry, installation snippet, and
  'Why git-remote-pr?' community section
- Deterministic validation in validate-skill-templates.ps1 for git-remote-pr
  integration
- CHANGELOG.md documentation of v0.11.0 release with all changes and
  impacts
Restructure agent-smith with operating model focus, add two new reference guides (agentic-engineering, automation), and update skill descriptions across all repo-managed skills to clarify intent and trigger precision. Update eval definitions for agent-smith.
Add visual hero images for eight skills to support documentation rendering and skill identification.
Update README skill table to reflect agent-smith's operating model for durable software and agentic engineering, with concise decision-oriented framing and updated purpose/rationale section.
Add test-agent-smith.ps1 validation suite and update validate-skill-templates.ps1 to support agent-smith testing infrastructure.
@gimlichael gimlichael self-assigned this Sep 23, 2026
@greptile-apps

greptile-apps Bot commented Sep 23, 2026 •

Copy link
Copy Markdown

RetriggerConfidence Score: 5/5

The PR appears safe to merge; no outstanding previous findings or actionable new defects remain.

Summary

This PR introduces the deterministic git-remote-pr workflow, expands repository guidance for PR routing, scratch isolation, and Markdown formatting, restructures agent-smith around a four-level operating model, and updates validation, evaluation, documentation, and visual assets across the skill catalog. The changes since the previous review replace a source-text assertion with a behavioral regression that injects a late cleanup failure and verifies that completed remote writes remain accurately reported.

  • Adds prepare, plan, execute, approval-integrity, and post-write verification tooling for GitHub pull requests.
  • Enforces exact upstream branch matching, complete committed-change coverage, and explicit approval boundaries.
  • Adds deterministic validation and regression suites for git-remote-pr and agent-smith.
  • Updates repository and skill documentation to use continuous-line Markdown prose and isolated scratch workspaces.
  • Adds hero images and catalog metadata for multiple skills.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A[Committed base-to-head changeset] --> B[Prepare deterministic evidence]
  B --> C[Generate title, body, and plan]
  C --> D[Render approval-bound preview]
  D --> E{Approved and state unchanged?}
  E -->|No| F[Invalidate transaction]
  E -->|Yes| G[Push branch if required]
  G --> H[Create or update pull request]
  H --> I[Verify metadata, head SHA, and files]
  I --> J[Report final or partial remote state]
Loading

Reviews (10) · Last reviewed commit: "🔒️ add late cleanup failure injection t..."

Comment thread skills/git-remote-pr/scripts/execute-pr.ps1
Comment thread skills/git-remote-pr/scripts/prepare-pr.ps1 Outdated
Comment thread scripts/tests/test-agent-smith.ps1 Outdated
Add upstream tracking branch validation to prevent PR creation or update when local and upstream branch names diverge after rename or when tracking is missing. The skill now validates that the tracked upstream branch matches the local branch name (accounting for allowed separator differences), exists on the configured remote, and fails closed with clear remediation guidance. This prevents silent misconfigurations where a renamed local branch still tracks an old upstream, which could cause the PR to target the wrong remote branch.
Update SKILL.md to document that missing or mismatched upstream tracking blocks both normal and yolo/auto modes. Add guidance on using git push --set-upstream or git branch --set-upstream-to to fix tracking before retrying. Clarify that the workflow discovers the upstream head branch and validates it matches the local branch name before attempting any remote operations.
Add test cases and assertions for upstream tracking validation. Include tests for missing upstream tracking, branch name mismatches after rename, and equivalent branch name normalization. Add eval case 15 to verify that renamed local branch with stale upstream tracking is detected and rejected with remediation guidance. Update existing eval expectations to clarify that only already-tracked branches are pushed.
Update README to communicate that git-remote-pr now validates upstream tracking and blocks when local and upstream branch names mismatch or when tracking is missing. Clarify that the skill discovers the upstream head branch and ensures it matches the current branch before proceeding.
Consolidated decision framework from 5 to 4 precedence levels. Merged repository-family conventions into local repository policy. Strengthened guidance to require concrete repository evidence when importing conventions. Updated evaluation criteria to reflect these changes.
Updated skill instructions and evaluation criteria for git-remote-pr.
Refactored policy validation from sequential string search to regex-based extraction. Added assertions for delivery-and-repositories.md content validation and removed repository-family terminology checks. Updated test scenarios 20 and 30.
Updated README with current evaluation infrastructure, skill installation, and synchronization guidance.
Comment thread skills/git-remote-pr/scripts/pr-common.ps1 Outdated
The README design principles section now includes guidance on keeping diff reviews bounded with git diff --stat, path-specific inspection, and verified temporary storage use only as a cleanup-verified fallback.
Reorganized the Working-tree Scratch Isolation section to emphasize bounded diff review: starting with git diff --stat, using path-specific inspection, and using temporary storage only as a cleanup-verified fallback. Updated eval #29 expectations to enforce this approach for step 1 diff inspection.
Enhanced test-agent-smith.ps1 to support optional Ref parameter for historical regression testing, materializing repair scripts in isolated temporary directories. Added new test-agent-smith-ref.ps1 for comprehensive regression isolation validation. Integrated ref isolation tests into validate-skill-templates.ps1 workflow.
aicia-bot and others added 11 commits September 24, 2026 00:03
Updated README validation documentation to clarify that execution refuses plans if approved title, ready/draft state, body, or evidence changed after approval. Added draft state to post-write verification checks.
Added approval integrity validation that detects if title or draft state changed between plan approval and execution. New Get-PlanApprovalKey function computes SHA256 hash of title/draft state for integrity checking. Removed branch name normalization to enforce exact branch name matching. Updated tests to verify tampering detection and strict branch equivalence.
Updated Step 6 guidance in git-remote-pr SKILL.md with detailed presentation rules for PR body synthesis: opening paragraph explaining purpose and outcomes, bold thematic headings with trailing colons, and compact outcome-oriented bullets rather than file-by-file inventory. Applied density guidelines (3-6 sections, approximately 8-12 bullets) as guidance rather than quotas. Updated related eval case expectations to match the new format guidance, including emphasis on bold-heading-colon syntax and outcome-collapsed presentation.
Updated the 'Why git-remote-pr?' section in README.md to clarify that the skill collapses evidence into theme-based descriptions with opening paragraph, bold thematic headings, and concise bullets—not file-by-file catalog. Explains how related helpers, validation cases, references, formatting changes, and assets collapse into shared reviewer-facing outcomes. Emphasizes that complete evidence coverage does not require complete prose enumeration and that repository templates retain precedence.
Updated CHANGELOG.md with comprehensive v0.11.0 release notes documenting the introduction of git-remote-pr with upstream tracking validation and PR approval integrity checking, refinements to git-visual-commits scratch isolation guidance, restructuring of agent-smith around an operating model framework with reduced decision precedence levels, comprehensive markdown prose formatting standardization across 12 skills, enhanced skill validation suites including regression testing support, and eight hero images for visual identification. Includes detailed Added, Changed sections covering new capabilities, modified workflows, and integration enhancements across the repository.
Refined the git-remote-pr skill workflow with improved guidance on PR preparation, evidence collection, and completion criteria. Enhanced eval cases to ensure workflow coverage. Updated PR implementation scripts for better error handling, theme validation, and interactivity support. Synced README skill description to match the current implementation.
The git-remote-pr skill now validates PR body format before creating or updating pull requests. The validator enforces exactly one opening paragraph, bold colon-ended themes, dash bullets under every theme, and rejects redundant summary sections.

Add Assert-PrBodyStructure function to pr-common.ps1 with comprehensive structure validation. Integrate the validator into make-plan.ps1 so malformed bodies fail early. Add test-pr-body.ps1 with 28 test cases covering valid and invalid body structures.
Expand test-pr.ps1 with comprehensive test coverage for the new body structure validation. Add upfront test that verifies Assert-PrBodyStructure passes and fails as expected. Add integration tests for malformed bodies that bypass planning and don't create plans or make GitHub writes. Add recovery tests to ensure regenerated valid bodies work correctly in the same workspace. Add template-override tests to validate that body validation respects repository templates and covers all files under template themes.
Update evals.json with new test cases covering PR body structure validation. Add cases that verify the planner rejects malformed bodies and accepts well-formed ones. Ensure evaluation confirms that body validation integrates correctly with the workflow and respects both default and template-provided body structures.
Update the git-remote-pr skill description in README.md to document the new body structure validation. The description now explains that the planner enforces exactly one opening paragraph, bold colon-ended themes, dash bullets under every theme, and rejects concluding summary sections. This change makes the README reflect the current skill behavior and helps users understand the validation rules.
Comment thread skills/git-remote-pr/scripts/make-plan.ps1
Implement preview artifact binding and integrity checking to ensure the exact preview.md shown to the user cannot be modified between planning and approval. Adds preview_hash field to plan object and validates preview file integrity before executing PR writes.
Update skill instructions and evaluation definitions to document preview artifact hash checking before any push or PR metadata write.
Add test assertions to verify preview hash matches file integrity and detect preview tampering.
Minor documentation updates to readme.
Implement approval validation system that detects and prevents PR execution when repository state drifts between preview generation and execution. Adds approval ID generation, invalidation tracking, and comprehensive drift diagnostics.
Update skill instructions and evaluation definitions to document approval validation and drift detection capabilities.
Add comprehensive test cases covering approval invalidation scenarios, state drift detection, and diagnostics.
Document approval validation and drift detection safety features in project README.
Comment thread skills/git-remote-pr/scripts/execute-pr.ps1 Outdated
aicia-bot and others added 2 commits September 24, 2026 21:07
Enhance the approval workflow to bind the preview more tightly to the approval ID. Normalize only the two generated approval-ID fields before hashing to compute review_hash, then derive the approval ID from both the write intent and review_hash. This prevents approval drift where the preview could be altered after approval without invalidating the transaction. During execution, verify review_hash by normalizing those fields back, ensuring the exact preview text matches what was approved.
Comment thread skills/git-remote-pr/scripts/test-pr.ps1 Outdated
@gimlichael
gimlichael merged commit 9975dda into main Sep 24, 2026
14 checks passed
@gimlichael
gimlichael deleted the v0.11.0/chore-and-git-remote-pr branch September 24, 2026 20:19
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