ilo-lang · danieljohnmorris · May 20, 2026 · May 20, 2026 · May 20, 2026 · May 19, 2026
diff --git a/.claude-plugin/marketplace.json b/.claude-plugin/marketplace.json
@@ -11,15 +11,15 @@
       "name": "ilo",
       "source": "./",
       "description": "Write, run, debug, and explain programs in ilo — a token-optimised programming language for AI agents",
-      "version": "0.12.0",
+      "version": "0.13.0",
       "author": {
         "name": "Daniel Morris"
       },
       "homepage": "https://github.com/ilo-lang/ilo",
       "license": "MIT",
       "keywords": ["ilo", "programming-language", "token-optimised", "ai-agents"],
       "skills": [
-        { "name": "ilo-language", "description": "Use this when writing or reviewing .ilo source.", "path": "skills/ilo/ilo-language.md" },
+        { "name": "ilo-language", "description": "Use this when writing or reviewing .@ source (.ilo also accepted with deprecation warning).", "path": "skills/ilo/ilo-language.md" },
         { "name": "ilo-language-records", "description": "Use this when writing ilo code that declares or uses record types.", "path": "skills/ilo/ilo-language-records.md" },
         { "name": "ilo-builtins-core", "description": "Use this when calling core builtins: type coercions, list ops, HOFs, and map ops.", "path": "skills/ilo/ilo-builtins-core.md" },
         { "name": "ilo-builtins-math", "description": "Use this when calling math builtins: arithmetic, trig, constants, random, and statistics.", "path": "skills/ilo/ilo-builtins-math.md" },

diff --git a/.github/gitleaks.toml b/.github/gitleaks.toml
@@ -5,6 +5,9 @@
 # an allowlist for the placeholder strings that ship in examples/ so the
 # release-gate scanner does not false-positive on demo code.
 #
+# For the full release-gate runbook (when to run, incident procedure, why
+# release-only) see docs/release-secret-scan.md.
+#
 # Run locally:
 #   gitleaks detect --source . --no-git --redact --verbose
 #   gitleaks detect --source . --redact --verbose   # includes git history

diff --git a/.github/workflows/sync-next.yml b/.github/workflows/sync-next.yml
@@ -17,15 +17,21 @@ jobs:
           ref: next
           fetch-depth: 0
           token: ${{ secrets.GITHUB_TOKEN }}
-      - name: Try fast-forward
+      - name: Try fast-forward, then 3-way merge
         id: ff
         run: |
           git config user.name "github-actions[bot]"
           git config user.email "github-actions[bot]@users.noreply.github.com"
           if git merge --ff-only origin/main; then
             echo "result=ff" >> "$GITHUB_OUTPUT"
             git push origin next
+          elif git merge --no-ff origin/main -m "chore(sync): merge main into next [automated]"; then
+            # Non-ff but clean 3-way merge: push the merge commit to next.
+            # Only falls through to the PR step on true conflict.
+            echo "result=merged" >> "$GITHUB_OUTPUT"
+            git push origin next
           else
+            git merge --abort || true
             echo "result=diverged" >> "$GITHUB_OUTPUT"
           fi
       - name: Open sync PR if diverged

diff --git a/.gitignore b/.gitignore
@@ -20,6 +20,9 @@ research/*
 research/closed-loop-bench/results/raw/
 research/closed-loop-bench/.zero/
 research/closed-loop-bench/**/__pycache__/
+# Zero compiler local cache (created when `zero check`/`zero build` runs
+# from the repo root, e.g. via Stage 5e tests). Reproducible; never tracked.
+/.zero/
 site/
 .DS_Store
 .astro/

diff --git a/.zero-version b/.zero-version
@@ -0,0 +1 @@
+0.1.2
diff --git a/CHANGELOG.md b/CHANGELOG.md
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
@@ -2,6 +2,8 @@
 
 Thanks for your interest in contributing to ilo!
 
+For the release secret-scan gate (gitleaks, allowlist, incident procedure) see [docs/release-secret-scan.md](docs/release-secret-scan.md).
+
 ## Getting Started
 
 ```bash

diff --git a/Cargo.lock b/Cargo.lock
diff --git a/Cargo.toml b/Cargo.toml
@@ -1,6 +1,6 @@
 [package]
 name = "ilo"
-version = "0.12.0"
+version = "0.13.0"
 edition = "2024"
 rust-version = "1.85"
 description = "ilo - the token-minimal programming language AI agents write"
@@ -46,6 +46,8 @@ fastrand = "2"
 getrandom = "0.2"
 regex = "1"
 chrono = { version = "0.4", default-features = false, features = ["clock"] }
+wasm-encoder = "0.249"
+tempfile = "3"
 percent-encoding = "2"
 base64 = "0.22"
 chrono-tz = "0.10"
@@ -57,8 +59,9 @@ subtle = "2"
 [dev-dependencies]
 wiremock = "0.6"
 tokio = { version = "1", features = ["rt-multi-thread", "macros"] }
-tempfile = "3"
 serde_json = "1"
+wasmparser = "0.249"
+shlex = "1"
 
 [profile.release]
 strip = true

diff --git a/MANIFESTO.md b/MANIFESTO.md
@@ -29,6 +29,8 @@ A named argument like `amount: 42` costs more tokens than positional `42`. We in
 **What the agent cares about:** "How many tokens will this cost me end-to-end?"
 **How this helps:** The language is as terse as possible *without increasing retry rate*. Where there's a tradeoff between generation cost and error rate, we optimise for total cost.
 
+We measured the tokenizer; `.@` saves one token per filename.
+
 **Prefix notation** eliminates parentheses and saves tokens at every nesting level. `(a * b) + c` becomes `+*a b c` - 4 fewer characters, 1 fewer token. Deeper nesting saves more: `((a + b) * c) >= 100` becomes `>=*+a b c 100` - 7 fewer characters, 3 fewer tokens. Across 25 expression patterns, prefix notation saves 22% of tokens and 42% of characters vs infix. See the [prefix-vs-infix benchmark](research/explorations/prefix-vs-infix/) for the full analysis.
 
 **Guards instead of if/else** eliminate nesting depth. In a traditional language, conditional logic stacks:

diff --git a/README.md b/README.md
@@ -104,20 +104,21 @@ Uses the [skills](https://www.npmjs.com/package/skills) npm package (396K+ insta
 # Inline
 ilo 'dbl x:n>n;*x 2' 5                    # → 10
 
-# From file
-ilo program.ilo functionName arg1 arg2
+# From file (.@ and .ilo are both accepted; .@ saves one token per filename on LLM tokenizers)
+ilo program.@ functionName arg1 arg2
+ilo program.ilo functionName arg1 arg2   # .ilo works too
 
 # Verb form (cargo / go / zero style; bare positional still works)
-ilo run program.ilo arg1 arg2             # run
-ilo check program.ilo                     # verify only - exit 0 if clean
-ilo build program.ilo -o ./bin            # AOT compile
+ilo run program.@ arg1 arg2              # run
+ilo check program.@                      # verify only - exit 0 if clean
+ilo build program.@ -o ./bin             # AOT compile
 ```
 
 **[Tutorial: Write your first program →](https://ilo-lang.ai/docs/first-program/)**
 
 ## Editor support
 
-Syntax highlighting, snippets, and `--` comment handling for `.ilo` files ships in [`extensions/vscode/`](./extensions/vscode/). Install into Cursor with `cd extensions/vscode && npm run install:cursor`. VS Code marketplace publish is tracked separately.
+Syntax highlighting, snippets, and `--` comment handling for `.ilo` and `.@` files ships in [`extensions/vscode/`](./extensions/vscode/). Install into Cursor with `cd extensions/vscode && npm run install:cursor`. VS Code marketplace publish is tracked separately.
 
 ## Versioning
 

diff --git a/SECURITY.md b/SECURITY.md
@@ -1,80 +1,10 @@
-# Release security
+# Security policy
 
-This page documents the security gates that run before an ilo release tag is
-cut. The goal is simple: a leaked credential, API key, private key, or other
-secret should never make it onto a published artifact, a published crate, a
-published npm package, or a GitHub release.
+Found a vulnerability in ilo? Please do not open a public issue.
 
-## Secret scan (gitleaks)
+Report it privately via [GitHub's private vulnerability reporting](https://github.com/ilo-lang/ilo/security/advisories/new).
 
-Every push of a `v*` tag triggers `.github/workflows/release.yml`. The first
-job is `secret-scan`, which runs
-[`gitleaks/gitleaks-action@v2`](https://github.com/gitleaks/gitleaks-action)
-over the full repository history. All downstream jobs (`build`, `build-wasm`,
-`release`, `publish-crates`, `publish-npm`, `publish-pi`) declare
-`needs: secret-scan`, so any finding blocks the entire release.
+We aim to acknowledge reports within 72 hours.
 
-### What gets scanned
-
-- Working tree (every tracked file).
-- Full git history (`fetch-depth: 0`).
-- Default gitleaks rule pack: AWS, GCP, Azure, GitHub, OpenAI, Anthropic,
-  Stripe, Slack, JWT, generic high-entropy strings, PEM blocks, and more.
-
-### Whitelist
-
-Placeholder credentials shipped in `examples/` (especially `examples/apps/*`
-for LLM-client and ScrapingBee demos) are explicitly allowed in
-[`.github/gitleaks.toml`](./.github/gitleaks.toml). The current allow regex set:
-
-- `SCRAPINGBEE_KEY_PLACEHOLDER_set_via_env_in_real_use`
-- `sk-PLACEHOLDER[-_A-Za-z0-9]*`
-- `REPLACE_ME` / `YOUR_*_HERE` / `EXAMPLE_*_KEY`
-
-If a new example needs a placeholder credential, add it to the allowlist in
-the same PR.
-
-### Running locally
-
-Before pushing a tag, or any time you want to sanity-check the working tree:
-
-```sh
-gitleaks detect --source . --no-git --redact --verbose
-gitleaks detect --source . --redact --verbose   # includes git history
-```
-
-A clean run prints `no leaks found`. Anything else is a real finding to
-triage before the release goes out.
-
-## Install-script integrity verification
-
-The release workflow's `release` job runs `sha256sum ilo-* > checksums-sha256.txt`
-and uploads the resulting file alongside every published binary. The
-`curl ... | sh` installers shipped from `https://ilo-lang.ai/install.sh` and
-`/install.ps1` (canonical source in [`scripts/install/`](./scripts/install/))
-fetch that checksum file together with the binary and refuse to install if
-the SHA-256 doesn't match. This closes the standard supply-chain attack
-window on the curl-pipe install path: a tampered binary on GitHub's CDN, a
-TLS-intercepted download, or a mirrored asset all fail the check before the
-binary is made executable. An offline regression test
-(`scripts/install/test-install-sh.sh`) runs on every CI push and exercises
-the happy, tamper, and missing-asset code paths.
-
-## Why release-only, not per-PR
-
-Running gitleaks on every PR added meaningful queue time without much
-incremental safety: secrets in feature branches are caught at merge time by
-GitHub's native push-protection, and the release gate is the last guarantee
-before anything becomes public. The release-only model keeps developer
-feedback fast and still blocks the public artifact path.
-
-## If the scan finds something
-
-1. The release job will fail with `secret-scan` red. No artifacts are built.
-2. Treat the finding as a real incident: rotate the credential immediately,
-   regardless of where the leak appears (working tree, history, comment, or
-   doc).
-3. Once rotated, scrub the secret from history (`git filter-repo` or
-   BFG), force-push the cleaned history, and re-cut the tag.
-4. If the finding is a false positive on a new placeholder shape, extend the
-   allowlist in `.github/gitleaks.toml` in a follow-up PR and re-cut the tag.
+For the release-time secret-scan gate, install-script integrity verification,
+and incident response procedure, see [`docs/release-secret-scan.md`](./docs/release-secret-scan.md).
-Original file line number
+Diff line change
@@ Expand Up / @@ -2,6 +2,8 @@ @@
     Thanks for your interest in contributing to ilo!
+    For the release secret-scan gate (gitleaks, allowlist, incident procedure) see [docs/release-secret-scan.md](docs/release-secret-scan.md).
     ## Getting Started
     ```bash
@@ Expand Down @@