v1.1.1 — OpenAI Codex CLI integration

Patch release adding OpenAI Codex CLI to the auto-integration set.

What's new

• --integrate=codex — global integration writing ~/.codex/AGENTS.md. One install covers every Codex session.
• Auto-detect — --integrate=auto picks up ~/.codex/ and wires Codex automatically alongside Cursor / Copilot / Antigravity / Zed.
• --integrate=all now includes codex (cursor + copilot + copilot-cli + antigravity + codex + zed).
• Per-repo Codex — already worked free via --integrate=antigravity-repo. Codex respects the cross-tool AGENTS.md standard, so the existing per-repo flag double-duties for it.

Why

Codex CLI is widely deployed and asked for. ChatGPT desktop and ChatGPT Agent (Operator) remain out of scope: desktop has no shell exec, Operator runs in a cloud sandbox with no local filesystem.

Compatibility

Additive only — no changes to SAMP wire format, JSONL schema, or existing integrations. All 53 tests green.

Upgrade

git pull && ./install.sh
./install.sh --integrate=codex     # or --integrate=auto to pick up everything

Docs

• Codex CLI integration guide
• All integrations

v1.1.0 — multi-agent integrations + perf

Minor release. Additive only — no breaking changes.

Highlights

• One-click multi-agent integrations. ./install.sh --integrate=<tool> writes a marker block so any agent CLI knows how to use the wrapper. Supported: cursor, copilot, copilot-cli, antigravity (global + antigravity-repo per-repo), zed. --integrate=auto detects what you have installed; --integrate=all enables every global integration.
• Wrapper mtime short-circuit. Idle inbox now ~5x faster on the Claude path (50k records: ~100ms → ~20ms). Both readers stat-and-skip before parsing; speedup compounds with the existing shell path.
• SAMP §3 / §5 / §6.6 hardening. Body NFC normalization for cross-impl id stability, reader-side enforcement of single-writer rule (records where from ≠ filename alias are dropped), atomic watermark writes via os.replace, O_NOFOLLOW on log writes + symlink filter on reads.
• samp-validate conformance tool. Standalone Python script (~144 lines) — point it at any \$DIR to verify writer-side SAMP §9 conformance. Catches tampered ids, single-writer violations, symlinked logs.
• --version flags on ~/.agent-message-cmd, msg, and samp-validate.
• Stdin EOF guard. send/reply now fail fast on interactive TTY instead of hanging. Pipe via `<`, `<<<`, or `echo … |`.
• CI: ruff + pyright job for Python; shellcheck on Ubuntu; round-trip suite on Ubuntu + macOS (51 tests).

Spec

No spec changes. Still SAMP v1 — see SPEC.md. All v1.0.0 logs and watermarks remain readable.

Implementations

See IMPLEMENTATIONS.md. Validate yours with `./samp-validate /path/to/$DIR`.

Upgrade

```bash
git pull && ./install.sh
```

Re-running install is idempotent. Verify: `~/.agent-message-cmd --version` should print `agent-message 1.1.0 (SAMP v1)`.

Full changelog: v1.0.0...v1.1.0

v1.0.0 — SAMP v1, first stable release

First stable release of SAMP (Simple Agent Message Protocol) — file-based, append-only messaging between AI agents. No server, no MCP, no daemon, no token.

Highlights

• Spec frozen at v1. See SPEC.md. §11 documents versioning policy: additive-only within v1.
• Three paths, one on-disk format:
  • Claude Code slash commands: /message-send, /message-inbox, /message-reply — 1 Bash tool call per op.
  • msg shell function — 0 LLM tokens, never touches a model.
  • Wrapper executable (~/.agent-message-cmd) — any agent CLI / framework / cron / script with a Bash or subprocess tool.
• Per-writer logs + content-addressed IDs. Sync conflict-free across machines via Syncthing / Dropbox / iCloud.
• mtime short-circuit in shell path keeps idle inbox latency at the python3 startup floor (~30ms).
• Default $DIR: ${XDG_STATE_HOME:-$HOME/.local/state}/agent-message/.

Implementations

Implementation	Language	Role
agent-message	Python + bash	both
agent-deck	Go	reader

See IMPLEMENTATIONS.md.

Install

git clone https://github.com/slima4/agent-message && cd agent-message && ./install.sh

Re-run is idempotent. ./install.sh --uninstall cleans up.

Docs

https://slima4.github.io/agent-message/