chore: GitHub issue + PR templates + CODEOWNERS

.github/CODEOWNERS

@@ -0,0 +1,5 @@
+# Code ownership. Solo-owner repo — expand when a second human joins the project.
+# See docs/DEVELOPMENT.md and memory project_solo_owner_admin_merge for
+# the review/merge workflow context.
+
+*                               @constk

.github/ISSUE_TEMPLATE/bug.md

@@ -0,0 +1,33 @@
+---
+name: Bug report
+about: Something broke — agent gave wrong output, an endpoint errored, a test regressed in prod-like conditions.
+title: "fix: "
+labels: ["bug"]
+---
+
+## What happened
+
+
+
+## What I expected
+
+
+
+## How to reproduce
+
+<!-- Concrete steps. Include the exact question/input for agent bugs. -->
+
+
+
+## Environment
+
+- Version / commit:
+- Browser + OS (if frontend):
+- LLM model / deployment (if agent):
+
+## Relevant invariant
+
+<!--
+  If this is a violation of a rule in docs/INVARIANTS.md, cite it by number.
+  Otherwise "none" — but consider whether this suggests a *missing* invariant.
+-->

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,2 @@
+# Block the "open a blank issue" option so everything gets structured.
+blank_issues_enabled: false

.github/ISSUE_TEMPLATE/eval-regression.md

@@ -0,0 +1,44 @@
+---
+name: Eval regression
+about: A golden-dataset question that used to pass now fails (often auto-filed by eval-nightly).
+title: "eval: regression on [case_id]"
+labels: ["bug", "test"]
+---
+
+## Case
+
+- **ID:**
+- **Category:**
+- **Difficulty:**
+
+## Question
+
+
+
+## Expected answer
+
+
+
+## Actual answer
+
+
+
+## When did it regress?
+
+<!-- First failing commit / run / model change / prompt edit. -->
+
+
+## Reproduction
+
+```bash
+uv run pytest eval/test_golden_qa.py -k "<case_id>" -v
+```
+
+## Suspected cause
+
+<!-- Prompt drift? Tool output shape change? Model upgrade? Data edit? -->
+
+
+## Invariant touched
+
+<!-- Does this also break a rule in docs/INVARIANTS.md? If so, cite it. -->

.github/ISSUE_TEMPLATE/feature.md

@@ -0,0 +1,25 @@
+---
+name: Feature request
+about: Propose a new capability, new tool, or architectural change.
+title: "feat: "
+labels: ["enhancement"]
+---
+
+## Problem
+
+<!-- What can't be done today, or what's painful about doing it? -->
+
+
+## Proposed solution
+
+<!-- Concrete: files touched, new contracts if any, APIs affected. -->
+
+
+## Acceptance criteria
+
+- [ ]
+- [ ]
+
+## Priority rationale
+
+<!-- Why now vs later? What's the cost of not doing this? -->

.github/pull_request_template.md

@@ -0,0 +1,45 @@
+<!--
+  Fill every section. "None" is a valid answer where applicable. Sections
+  you delete without answering will come back in review.
+-->
+
+## What & why
+
+<!-- 1-3 lines: what does this PR change, and why now? -->
+
+
+## Test plan
+
+<!-- Checkbox list. CI covers most of this; note any manual verification. -->
+
+- [ ]
+
+## Invariants affected
+
+<!--
+  Which rules from docs/INVARIANTS.md does this PR touch? Reference by
+  number (e.g. "#4 — bounded chat messages"). "None" is valid.
+-->
+
+
+## New deps / actions / external surface
+
+<!--
+  Anchor for the supply-chain review layers (Dependabot, pip-audit,
+  npm audit, Trivy, gitleaks, CodeQL-placeholder). Note any:
+  - new Python or npm packages (beyond what Dependabot updates auto-manage)
+  - new GitHub Actions uses: …@... lines (must be SHA-pinned for third-party)
+  - new external endpoints the app calls (LLM, database, webhook, etc.)
+  - new tools downloaded or installed at runtime
+  "None" is valid and common.
+-->
+
+
+## Screenshots
+
+<!-- Required for UI change. Delete this section for non-UI PRs. -->
+
+
+## Linked issue
+
+Closes #