docs(064): Glass Cockpit spec + agent-instruction bundles#557
Merged
Conversation
Spec, plan, and design artifacts for making the existing Paperclip "MCPProxy" cockpit (spec 045) transparent and steerable: invert the default from "proceed" to "checkpoint at every design-decision boundary" via three human gates (plan-of-attack, per-spec design, pre-merge) mapped to Paperclip native primitives (executionPolicy approval stages, request_confirmation/suggest_tasks interactions, issue_tree_holds), with reasoning visible before each gate and a single "waiting on you" view. Phased rollout: A) config + agent-instruction only (the dry-run target), B) a Paperclip plugin for the fused transparency UI, C) a fork only if A/B fall short. SynapBus is log/wiki only, never on the critical path. Includes rewritten gate-aware agent instructions, consumed-API + executionPolicy + agent-instruction contracts, data model, research, and operator quickstart. Supersedes spec 062's fresh-dev-instance approach; extends spec 045.
…n veto Session 2 amendment to FR-005 + US3: replace the mandatory human-merge gate (throughput bottleneck) with draft-PR + dual-AI-review consensus auto-merge. Two reviewers on different model families (Gemini 2.5-pro Critic + Codex), never the implementer; tests-green + both-accept → auto-merge; human is an optional 3rd reviewer with veto (request-changes/hold freezes auto-merge). Prerequisite flagged: a bot GitHub identity (agents currently = author's gh, and GitHub forbids self-approval) — interim fallback is 2-AI-review-as- required-check with the human merging. codex-local Paperclip adapter exists.
…raft-PR + setup Adds the Session-2 gate-model deliverables: reviewer/REVIEWER.md (shared RV-1..RV-6 dual-review doctrine), codex-reviewer/AGENTS.md (2nd reviewer on codex-local), amends engineer ENG-4 to open DRAFT PRs + request 2 AI reviewers (no self-merge), and auto-merge-setup.md (GitHub branch-protection config + the bot-identity prerequisite + interim human-merge fallback + open items).
…d, Codex gpt-5.5 ready Per user directive: both AI reviewers use paid SUBSCRIPTION logins, not API keys. - Gemini Critic: subscription/OAuth, pin gemini-2.5-pro (3.5/3 UNVERIFIED — quota exhausted on every probe today; switch if confirmed). TWO blockers: quota + empty-prompt adapter bug → cannot accept yet. - Codex reviewer: ChatGPT subscription, gpt-5.5 (codex-cli 0.46.0) — READY now. - Live two-reviewer set today = Codex + human (FR-005f) until Gemini recovers.
…Codex as live pair Gemini settings pin gemini-3.1-pro-preview (subscription/OAuth) but quota is exhausted (no reset hint) + empty-prompt adapter bug → Critic can't accept yet. Live 2-AI reviewer pair = Codex gpt-5.5 + Kimi-K2.5 (opencode_local, Gcore key present); Gemini rejoins as 3rd reviewer when quota returns.
Live dual-AI reviewer pair created in the running Paperclip cockpit and verified responding (2026-05-31): - CodexReviewer — codex_local / gpt-5-codex (5b94562c-…) - KimiReviewer — opencode_local / Kimi-K2.5 (fdaa1d4c-…) Both carry managed instruction bundles (shared doctrine + RV-1..RV-6 + role notes), report to CEO, idle, heartbeat off (woken by review-stage). Docs: - add canonical kimi-reviewer/AGENTS.md (the design lacked it) - correct codex-reviewer/AGENTS.md model facts (gpt-5.5 -> gpt-5-codex) - auto-merge-setup.md: live pair is Codex+Kimi; Gemini Critic becomes the 3rd reviewer when its subscription quota recovers codex config fix (~/.codex/config.toml, not in repo): model_reasoning_effort xhigh->high and model gpt-5.5->gpt-5-codex. On codex-cli 0.46.0 + ChatGPT subscription auth, gpt-5.5 needs a newer CLI and gpt-5.4/5.3-codex/5.2 are auth-restricted; gpt-5-codex/gpt-5 are the working models. Backup at ~/.codex/config.toml.bak.pre-reviewer-fix.*
engineer/AGENTS.md: - ENG-8: drive every required check to green before review — run local verification before push, watch `gh pr checks --watch`, push fixes until all green; never leave/hand off a red PR, never --no-verify or weaken a check. Green CI is the engineer's job, not the reviewer's. - ENG-9: when a change touches CLI/REST/MCP API/config/defaults/security or anything under docs/, the SAME PR must update docs/ (+ CLAUDE.md/ oas/swagger.yaml/README where mirrored). Docs-only changes exempt from TDD. - ENG-5 reworked to dual-AI merge-readiness (Codex+Kimi accept + all CI green). reviewer/REVIEWER.md RV-3: red/pending check = automatic request_changes; missing docs when the change warrants them = request_changes. Applied to the live Paperclip brains: 3 engineers (Backend/Frontend/MacOS) re-flattened from canonical; Codex+Kimi reviewer brains refreshed.
Phase-1 gate live on main (no bot identity needed): required_status_checks strict=false with 8 always-run, non-path-conditional contexts (Lint, Unit Tests ubuntu, Build ubuntu/macos/windows, Build Frontend, Validate PR title, Verify OpenAPI Artifacts). Existing 1-review + enforce_admins=false kept. Verified: green PR #553 satisfies all 8 (blocked only by review); in-flight PR #555 blocked on pending required checks. Documents the deliberately- excluded checks and the Go-version-pinned context-name fragility.
Deploying mcpproxy-docs with
|
| Latest commit: |
d10a31c
|
| Status: | ✅ Deploy successful! |
| Preview URL: | https://701b7cbd.mcpproxy-docs.pages.dev |
| Branch Preview URL: | https://064-glass-cockpit.mcpproxy-docs.pages.dev |
Forking a work branch from another feature branch drags its unmerged commits into the PR (root cause of spec-064 docs leaking into the MCP-770 race fix #556). ENG-3 now mandates fetch + branch from origin/main explicitly, in both the engineer bundle and the contract.
|
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
📦 Build ArtifactsWorkflow Run: View Run Available Artifacts
How to DownloadOption 1: GitHub Web UI (easiest)
Option 2: GitHub CLI gh run download 26738911936 --repo smart-mcp-proxy/mcpproxy-go
|
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.
Summary
Spec 064 artifacts only — 22 files, all under
specs/064-glass-cockpit/, no code.spec.md,plan.md,research.md,data-model.md,quickstart.md,checklists/contracts/— agent-instruction behavioral contract +execution-policy.schema.json+ Paperclip API notesagent-instructions/— canonical per-role instruction bundles (AGENTS.md/GEMINI.md/REVIEWER.md) consumed byscripts/apply-instructions.shto provision the running cockpit agents' managed bundlesWhy a separate PR
These spec docs were inadvertently bundled into #556 (MCP-770 race fix) because that branch was forked from
064-glass-cockpitinstead ofmain. #556 has been rebased ontomainand now carries only the 11 race-fix files. This PR lands the 064 artifacts on their own somainholds them as the durable canonical source — and so future fix branches cut frommainalready include them.No Go/Swift/CI changes; merge order vs #556 doesn't matter (disjoint file trees).