CI + AI manifest v2 + branch safety

CI workflow (ci.yml):
- cargo check, test, clippy, fmt on every push/PR to main
- Grammar change detection (SECURITY warning on ebnf/pest changes)
- Harvard invariant check (data rules must not reference control)
- Sentinel hash stability (sha256 of grammar files)

AI Manifest v2:
- Session resume instructions for new Claude sessions
- Points to conversation log, FAQs, design journal, papers
- Notes known math errors in tropical types paper
- References 007 dogfooding methodology

Branch protection requires GitHub Pro for private repos — CI workflow
serves as the safety net instead.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: hyperpolymath

.github/workflows/ci.yml

@@ -0,0 +1,68 @@
+# SPDX-License-Identifier: PMPL-1.0-or-later
+# 007 CI — runs on every push and PR
+# This is the safety net: tests + clippy + Harvard invariant checks
+
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+permissions:
+  contents: read
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5
+      - uses: dtolnay/rust-toolchain@4be9e76fd7c4901c61fb841f559994984270fce7
+        with:
+          toolchain: nightly
+          components: clippy, rustfmt
+      - uses: Swatinem/rust-cache@779680da715d629ac1d338a641029a2f4372abb5
+
+      - name: Check
+        run: cargo check --all-targets
+
+      - name: Test
+        run: cargo test
+
+      - name: Clippy
+        run: cargo clippy -- -D warnings
+
+      - name: Format check
+        run: cargo fmt -- --check
+
+  grammar-guard:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5
+
+      - name: Grammar change detection
+        run: |
+          if git diff --name-only HEAD~1 2>/dev/null | grep -qE '(grammar\.ebnf|grammar\.pest)'; then
+            echo "::warning::SECURITY: Grammar files changed in this commit. Manual review required."
+            echo "Grammar changes affect Harvard Architecture safety guarantees."
+            echo "Changed files:"
+            git diff --name-only HEAD~1 | grep -E '(grammar\.ebnf|grammar\.pest)'
+          fi
+
+      - name: Harvard invariant check
+        run: |
+          # Verify data_expr rules don't reference control constructs
+          if grep -n 'control_stmt\|control_expr\|control_block' crates/oo7-core/src/grammar.pest | grep -i 'data_'; then
+            echo "::error::HARVARD VIOLATION: data rules reference control constructs"
+            exit 1
+          fi
+          echo "Harvard invariant: OK"
+
+      - name: Sentinel hash stability
+        run: |
+          # Track grammar hash for sentinel verification
+          sha256sum spec/grammar.ebnf > /tmp/grammar_hash.txt
+          sha256sum crates/oo7-core/src/grammar.pest >> /tmp/grammar_hash.txt
+          echo "Grammar hashes:"
+          cat /tmp/grammar_hash.txt

0-AI-MANIFEST.a2ml

@@ -5,10 +5,27 @@
 ;; Read this file FIRST before any other file in this directory.
 
 (ai-manifest
-  (version "1.0")
+  (version "2.0")
   (project "007")
   (type "programming-language")
-  (status "specification")
+  (status "active-development")
+  (license "ALL-RIGHTS-RESERVED")
+  (last-major-session "2026-03-24")
+
+  (session-resume
+    "This repo had a MAJOR session on 2026-03-24. If resuming work:
+     1. Read docs/conversation-2026-03-24.md for full context
+     2. Read docs/faq/ for design decisions with attribution
+     3. Read docs/DESIGN-JOURNAL.adoc for philosophical grounding
+     4. Grammar is at v0.5.0 (21 sections, 6 invariants)
+     5. Type checker covers L1-3, L7 (linear), L9 (session), @budget
+     6. Evaluator has full agent runtime (spawn/send/receive/exchange)
+     7. Elixir/OTP codegen works: oo7 compile <file>
+     8. Agent API: oo7 agent-validate, agent-moves, agent-card
+     9. 207 tests, zero clippy warnings
+     10. Two arXiv paper drafts in docs/paper/
+     11. Tropical token types paper has KNOWN MATH ERRORS — needs correction
+     12. Use 007 dogfooding methodology (see memory/feedback_007_dogfood.md)")
 
   (classification
     (dual-use #t)