chore(workflow): replace Codex agent layout with Claude Code

The repository targeted Codex sessions. Move the agent contract, skills, and
session configuration to their Claude Code equivalents so a clean session at
the repository root loads the correct instructions.

- AGENTS.md becomes CLAUDE.md.
- .agents/skills/ becomes .claude/skills/, dropping the Codex-only
  agents/openai.yaml interface metadata that Claude Code does not read.
- .codex/config.toml becomes .claude/settings.json, pinning the same model
  and reasoning effort in Claude Code's native format.

Tooling follows the layout rather than only its names: verify_metadata.py
drops interface_metadata() and TOML parsing and now validates the settings
JSON, and verification_scope.py drops the openai.yaml lightweight-path rule.
.claude/settings.json classifies as a control file, so editing it still fails
closed to the full hosted matrix, while SKILL.md edits stay lightweight.

Verification: tools/verify_metadata.py passes (5 skills, 19 links);
tests/tools/verify_metadata_test.py and
tests/tools/verification_scope_test.py pass (3 tests each).

Author: kaikisegfault

.agents/skills/change-protocol/agents/openai.yaml

@@ -0,0 +1,4 @@
+{
+  "model": "opus[1m]",
+  "effortLevel": "xhigh"
+}

.agents/skills/conclude-project/agents/openai.yaml

@@ -12,7 +12,7 @@ description: Design or implement consensus-visible changes in protocol-stack, in
 2. Classify the change: primitive, encoding, validation, state transition,
    economics, authority, compatibility, or adapter-only.
 3. Research credible alternatives, select a recommended default under the
-   standing delegation in `AGENTS.md`, and record the rationale and
+   standing delegation in `CLAUDE.md`, and record the rationale and
    consequences in an ADR. Do not pause for owner approval.
 4. Write or update the canonical specification before implementation.
 

.agents/skills/proceed-project/agents/openai.yaml

@@ -6,12 +6,12 @@ description: Conclude active work in protocol-stack when the owner says "conclud
 # Conclude the project session
 
 Treat conclusion as a delivery workflow, not a summary request. Finish all
-in-scope work already started and leave the repository safe for a clean Codex
-session. Do not start the next roadmap slice, even when it is unblocked.
+in-scope work already started and leave the repository safe for a clean Claude
+Code session. Do not start the next roadmap slice, even when it is unblocked.
 
 ## 1. Freeze and reconstruct scope
 
-1. Read `AGENTS.md`, `docs/project/current-state.md`,
+1. Read `CLAUDE.md`, `docs/project/current-state.md`,
    `docs/project/founder-constitution.md`, `docs/project/charter.md`, and
    `docs/project/first-goal.md`.
 2. Fetch and prune `origin`, verify GitHub authentication, and inspect the

.agents/skills/status-project/agents/openai.yaml

@@ -1,11 +1,11 @@
 ---
 name: proceed-project
-description: Continue autonomous work in protocol-stack when the owner says "proceed", "continue", or asks to resume from a clean Codex session. Reconstruct verified repository state, execute the next unblocked milestone slice, verify it, update the handoff, and keep working while safe work and execution capacity remain.
+description: Continue autonomous work in protocol-stack when the owner says "proceed", "continue", or asks to resume from a clean Claude Code session. Reconstruct verified repository state, execute the next unblocked milestone slice, verify it, update the handoff, and keep working while safe work and execution capacity remain.
 ---
 
 # Proceed with the project
 
-1. Read `AGENTS.md`, `docs/project/current-state.md`,
+1. Read `CLAUDE.md`, `docs/project/current-state.md`,
    `docs/project/founder-constitution.md`, `docs/project/charter.md`, and
    `docs/project/first-goal.md`.
 2. Read only the roadmap, ADRs, specifications, and engineering documents
@@ -36,7 +36,7 @@ description: Continue autonomous work in protocol-stack when the owner says "pro
     execution capacity remains. A completed slice is not a reason to yield.
 
 Use meaningful issues, focused branches, and evidence-bearing PRs as defined in
-`AGENTS.md`. Do not manufacture activity with empty commits or vanity issues.
+`CLAUDE.md`. Do not manufacture activity with empty commits or vanity issues.
 
 The repository's standing delegation makes ordinary owner approval
 unnecessary. Do not pause for technical, product-mechanism, protocol,
@@ -49,7 +49,7 @@ resolution of conflicting unexplained user work. Complete all other unblocked
 work first.
 
 Never use chat history as the authoritative handoff. Never claim continuous
-background execution after the current Codex run ends. Never leave detached
+background execution after the current Claude Code run ends. Never leave detached
 local work running. At each phase boundary, cancel obsolete GitHub runs, audit
 local processes, and remove reproducible artifacts with
 `tools/clean-local.sh`.

.agents/skills/verify-project/agents/openai.yaml

@@ -10,7 +10,7 @@ workflow. Do not create issues, branches, commits, PRs, or documentation edits.
 
 ## Reconstruct the evidence
 
-1. Read `AGENTS.md`, `docs/project/current-state.md`,
+1. Read `CLAUDE.md`, `docs/project/current-state.md`,
    `docs/project/founder-constitution.md`, `docs/project/charter.md`,
    `docs/project/first-goal.md`,
    `docs/project/roadmap.md`, `docs/project/vision.md`, and `README.md`.