Stop-the-line for PR #606 (memory) — no merges to memory-touching files until landed #903

Goal

Document and enforce a stop-the-line rule: no merges to memory-touching files until PR #606 lands. With coding agents working in parallel, every merge to main makes the 38k-LOC memory PR #606 staler and less mergeable. Without this rule, #606 becomes unmergeable and the consumer launch slips by 4+ weeks.

Why this matters

PR #606 (memory v2) is currently:

• 38,759 additions / 1,951 deletions across 77 files
• DRAFT, currently CONFLICTING with main
• Last updated April 20 (6 days ago at filing)
• The foundation for v0.20.0's 11 consumer-critical items

Every parallel agent merging to main increases the conflict surface. With agents potentially producing 10+ PRs/day across the consumer-critical track, without explicit stop-the-line discipline, #606 dies.

The rule

Until PR #606 lands:

Frozen file paths (no PRs may merge changes to)

• src/gaia/agents/base/agent.py
• src/gaia/agents/base/memory*.py (all files matching)
• src/gaia/agents/base/discovery.py
• src/gaia/ui/routers/memory.py
• src/gaia/apps/webui/src/pages/MemoryDashboard.tsx
• Anything feat(memory): agent memory v2 — second brain with hybrid search, LLM extraction, and observability dashboard #606 explicitly touches per its file list (see PR description)

Allowed

• PRs that touch ONLY non-frozen file paths (most v0.18.2 mobile work, Telegram, scheduler, MCP catalogue)
• PRs that feat(memory): agent memory v2 — second brain with hybrid search, LLM extraction, and observability dashboard #606 explicitly identifies as compatible additive work
• Documentation changes
• CI / test infrastructure changes
• Hotfixes to v0.17.5 / v0.18.0

Process

• Add stop-the-line label to PR feat(memory): agent memory v2 — second brain with hybrid search, LLM extraction, and observability dashboard #606
• Add a CODEOWNERS-enforced rule that PRs touching frozen paths require approval from the feat(memory): agent memory v2 — second brain with hybrid search, LLM extraction, and observability dashboard #606 author + a release manager
• Pin a banner comment on feat(memory): agent memory v2 — second brain with hybrid search, LLM extraction, and observability dashboard #606 listing the frozen paths and the rule
• Update AGENTS.md (filed alongside) with the rule + how to check before opening a PR

Communication to coding agents

Add to AGENTS.md and orchestration playbook (A1):

Before opening any PR, run gh pr list --label stop-the-line to check for active stop-the-line PRs. If any exist, verify your PR doesn't touch their frozen paths. If it does, comment on the stop-the-line PR asking when it's expected to land, and either (a) wait, or (b) carve out the non-frozen subset of your work to ship first.

Active focus on landing #606

In parallel with the freeze, the team prioritizes:

• One agent owns rebase + conflict resolution (continuous, not one-shot)
• One agent monitors CI and addresses test failures as they arrive
• One human / orchestrator agent shepherds review (assigns reviewers, chases comments, resolves discussions)
• Daily check-in on claudia_get_task_status for the rebase agent

Target: PR #606 lands by May 5 (one week from filing). If it slips beyond May 12, escalate — at that point, v0.20.0's June 2 ship date is at risk and we need to discuss either descope or schedule slip.

Why not just merge #606 immediately?

• 77-file PRs deserve careful review, especially when they're foundational
• Skipping review for speed creates worse outcomes than a 1-week careful review
• The freeze keeps the PR mergeable while review proceeds; skipping the freeze requires trading review depth for merge velocity, which is the wrong trade for foundation code

Generalizing — when to use stop-the-line in the future

• PR is > 5,000 LOC AND touches > 20 files AND is foundational (memory, agent base, schema)
• Multiple downstream PRs depend on it shipping
• Active parallel work in adjacent file trees
• Failure to land → calendar quarter slip

This is the canonical example. Document it in A1 (playbook) so future similar PRs trigger the same discipline.

Deliverables

• Frozen-paths list pinned to PR feat(memory): agent memory v2 — second brain with hybrid search, LLM extraction, and observability dashboard #606 as a comment
• stop-the-line label created and applied to feat(memory): agent memory v2 — second brain with hybrid search, LLM extraction, and observability dashboard #606
• AGENTS.md updated with the rule + check command
• CODEOWNERS update if branch protection allows (to enforce automatically)
• A1 (playbook) cross-references this issue as the canonical example

Acceptance criteria

• PR feat(memory): agent memory v2 — second brain with hybrid search, LLM extraction, and observability dashboard #606 lands by May 5 (target) or May 12 (latest acceptable)
• No PR merges to a frozen path between this issue's filing and feat(memory): agent memory v2 — second brain with hybrid search, LLM extraction, and observability dashboard #606's land date
• Coding agents reading AGENTS.md will check stop-the-line label before opening PRs

Dependencies

• Blocks: every consumer-critical v0.20.0 item (they all depend on feat(memory): agent memory v2 — second brain with hybrid search, LLM extraction, and observability dashboard #606's foundation)
• Adjacent: A1 (orchestration playbook documents stop-the-line as a pattern), AGENTS.md
• Targets: in effect immediately upon filing