[codex] Expand ANCP language adapter coverage

.gitignore

@@ -1,6 +1,7 @@
 # Generated verification and downloaded source snapshots
 verification-report.json
 /.ancp/
+**/.ancp/
 /research/source-docs/snapshots/
 /research/source-docs/fetch-report.json
 /research/source-docs/index.md

CHANGELOG.md

@@ -14,7 +14,7 @@ Initial public release.
 - Python reference package with `ancp` CLI.
 - Compiler-facing proxy entry points such as `ancp-cargo`, `ancp-tsc`, `ancp-python`, `ancp-kotlinc`, and `ancp-julia`.
 - Native-name shim installer via `ancp install-shims`.
-- Native-tool adapters for Python, TypeScript, JavaScript, Rust, Go, C/C++, Java, Kotlin, C#/.NET, Swift, Zig, Ruby, PHP, Dart, Scala, and Julia.
+- Native-tool adapters for Python, TypeScript, JavaScript, Rust, Go, C/C++, Java, Kotlin, C#/.NET, Swift, Zig, Ruby, PHP, Dart, Scala, Julia, Shell, PowerShell, Lua, Perl, R, Haskell, OCaml, Erlang, Elixir, Clojure, Nix, Terraform, Dockerfile, SQL, JSON, TOML, and YAML.
 - Markdown renderer for compact agent-facing diagnostic summaries.
 - Multilingual intentionally broken bug corpus.
 - Repository validation, contract audit, source fetch, and bug corpus scripts.

MANIFEST.in

@@ -10,8 +10,10 @@ recursive-include schemas *.json
 recursive-include taxonomies *.json
 recursive-include examples *.json
 recursive-include examples/buggy *
+recursive-exclude examples/buggy .ancp/* native-output*.txt
 recursive-include docs *.md
 recursive-include spec *.md
+recursive-include tools *.py
 include research/README.md
 include research/tooling-matrix.md
 include research/source-docs/sources.json

README.md

@@ -6,7 +6,7 @@ The core idea is simple:
 
 > Existing tools may keep their native internals. ANCP defines the stable machine contract that adapters expose to agents.
 
-ANCP turns tool output into structured diagnostics, repair hints, repair plans, verification steps, code graph facts, effect/capability metadata, and version-matched agent guidance. It is designed to sit above current languages such as TypeScript, Python, Rust, Go, Java, C, C++, C#, Swift, Kotlin, Zig, Ruby, PHP, and others.
+ANCP turns tool output into structured diagnostics, repair hints, repair plans, verification steps, code graph facts, effect/capability metadata, and version-matched agent guidance. It is designed to sit above current languages such as TypeScript, Python, Rust, Go, Java, C, C++, C#, Swift, Kotlin, Julia, Zig, Ruby, PHP, Lua, Perl, R, Haskell, OCaml, Erlang, Elixir, Clojure, Nix, Terraform, Dockerfile, SQL, JSON, TOML, YAML, and others.
 
 ## What This Repository Contains
 
@@ -30,6 +30,7 @@ This repository is the public ANCP 1.0 contract and Python reference implementat
 | Native-tool adapters | Implemented | [src/ancp/adapters](src/ancp/adapters) |
 | Bug corpus | Implemented | [examples/buggy](examples/buggy) |
 | Repository verifier | Implemented | [tools/verify_repo.py](tools/verify_repo.py) |
+| Toolchain availability checker | Implemented | [tools/check_toolchains.py](tools/check_toolchains.py) |
 
 ## Why ANCP Exists
 
@@ -165,6 +166,9 @@ julia app.jl
 kotlinc Main.kt
 gcc -fsyntax-only main.c
 clang++ -fsyntax-only main.cpp
+bash -n script.sh
+pwsh -NoProfile -File script.ps1
+terraform validate
 ```
 
 The native compiler output and exit code are preserved. ANCP writes a structured sidecar:
@@ -247,6 +251,7 @@ python tools/audit_contracts.py
 python tools/verify_repo.py
 pytest
 python tools/run_bug_corpus.py
+python tools/check_toolchains.py
 python -m build
 python -m twine check dist/*
 ```

RELEASE_NOTES.md

@@ -9,7 +9,8 @@ ANCP 1.0.0 ships the Agent Native Compiler Protocol contract and Python referenc
 - Diagnostic, repair, and effect taxonomies.
 - Reference CLI: `ancp`.
 - Compiler-name shims for invisible compiler-layer usage.
-- Native-tool adapters for Python, TypeScript, JavaScript, Rust, Go, C/C++, Java, Kotlin, C#/.NET, Swift, Zig, Ruby, PHP, Dart, Scala, and Julia.
+- Native-tool adapters for Python, TypeScript, JavaScript, Rust, Go, C/C++, Java, Kotlin, C#/.NET, Swift, Zig, Ruby, PHP, Dart, Scala, Julia, Shell, PowerShell, Lua, Perl, R, Haskell, OCaml, Erlang, Elixir, Clojure, Nix, Terraform, Dockerfile, and SQL.
+- Built-in parser adapters for JSON, TOML, and YAML.
 - Compact Markdown rendering for agent context.
 - Multilingual broken-code corpus for adapter smoke testing.
 - CI, packaging, verification scripts, and OSS release checklist.

docs/final-readiness-report.md

@@ -13,9 +13,9 @@ ANCP now contains:
 - diagnostic, repair, and effect taxonomies,
 - a Python package named `ancp`,
 - the `ancp` CLI,
-- prefixed compiler proxies such as `ancp-cargo`, `ancp-tsc`, `ancp-python`, `ancp-kotlinc`, and `ancp-julia`,
+- prefixed compiler proxies such as `ancp-cargo`, `ancp-tsc`, `ancp-python`, `ancp-kotlinc`, `ancp-julia`, `ancp-pwsh`, `ancp-terraform`, and `ancp-sqlfluff`,
 - native-name compiler shims installed with `ancp install-shims`,
-- adapters for Python, TypeScript, JavaScript, Rust, Go, C/C++, Java, Kotlin, C#/.NET, Swift, Zig, Ruby, PHP, Dart, Scala, and Julia,
+- adapters for Python, TypeScript, JavaScript, Rust, Go, C/C++, Java, Kotlin, C#/.NET, Swift, Zig, Ruby, PHP, Dart, Scala, Julia, Shell, PowerShell, Lua, Perl, R, Haskell, OCaml, Erlang, Elixir, Clojure, Nix, Terraform, Dockerfile, SQL, JSON, TOML, and YAML,
 - compact Markdown rendering for agents,
 - a multilingual intentionally broken bug corpus,
 - research notes and source document fetch tooling,
@@ -29,23 +29,23 @@ The local release was checked with these passes.
 | --- | --- | --- |
 | Syntax | `python -m compileall -q src tests tools` | passed |
 | Unit tests | `pytest` | 14 passed |
-| Source corpus | `python tools/fetch_sources.py` | fetched 59/59 source documents |
+| Source corpus | `python tools/fetch_sources.py` | fetched 77/77 source documents |
 | Contract audit | `python tools/audit_contracts.py` | passed |
 | Repo validator | `python tools/verify_repo.py` | passed, wrote `verification-report.json` |
 | CLI JSON | `ancp manifest \| python -m json.tool` | passed |
 | CLI JSON | `ancp capabilities \| python -m json.tool` | passed |
 | CLI JSON | `ancp schema \| python -m json.tool` | passed |
 | Shim install | `ancp install-shims --dir .ancp/bin --force` | passed |
 | Shim smoke | normal `python -m py_compile bad.py` through shim | exit 1, ANCP sidecar status `failed`, 1 diagnostic |
-| Bug corpus | `python tools/run_bug_corpus.py` | emitted ANCP JSON/Markdown for all 16 cases |
+| Bug corpus | `python tools/run_bug_corpus.py` | emitted ANCP JSON/Markdown for all 33 cases |
 | Bug corpus validation | `ancp validate .ancp/bug-corpus` | passed |
 | Package build | `python -m build` | built wheel and sdist |
 | Package metadata | `python -m twine check dist/*` | both artifacts passed |
 
 Local tool availability during verification:
 
-- Installed and producing diagnostics: Python, Go, Julia.
-- Not installed or not available in this environment: TypeScript compiler, Rust/Cargo, GCC/Clang, Java, Kotlin, .NET, Swift, Zig, Ruby, PHP, Dart, Scala.
+- Installed and producing diagnostics: Python, Go, Julia, JSON, TOML, YAML, PowerShell.
+- Not installed or not available in this environment: TypeScript compiler, Rust/Cargo, GCC/Clang, Java, Kotlin, .NET, Swift, Zig, Ruby, PHP, Dart, Scala, ShellCheck/Bash, Lua, Perl, R, GHC, OCaml, Erlang, Elixir, clj-kondo, Nix, Terraform, hadolint, SQLFluff.
 - Missing native tools correctly report `tool_failed`; they do not pretend to pass.
 
 ## Scope Boundary

docs/github-release-plan.md

@@ -127,7 +127,7 @@ Highlights:
 - diagnostic, repair, and effect taxonomies
 - reference CLI
 - compiler-name shim installation
-- native-tool adapters for Python, TypeScript, JavaScript, Rust, Go, C/C++, Java, Kotlin, .NET, Swift, Zig, Ruby, PHP, Dart, Scala, and Julia
+- native-tool adapters for Python, TypeScript, JavaScript, Rust, Go, C/C++, Java, Kotlin, .NET, Swift, Zig, Ruby, PHP, Dart, Scala, Julia, Shell, PowerShell, Lua, Perl, R, Haskell, OCaml, Erlang, Elixir, Clojure, Nix, Terraform, Dockerfile, SQL, JSON, TOML, and YAML
 - validation and conformance tooling
 - GitHub Actions CI
 - research-backed language/tooling matrix

docs/invisible-compiler-layer.md

@@ -95,6 +95,23 @@ The reference implementation can create shims for:
 - `scala-cli`
 - `scalac`
 - `julia`
+- `shellcheck`
+- `bash`
+- `pwsh`
+- `powershell`
+- `luac`
+- `lua`
+- `perl`
+- `Rscript`
+- `ghc`
+- `ocamlc`
+- `erlc`
+- `elixirc`
+- `clj-kondo`
+- `nix-instantiate`
+- `terraform`
+- `hadolint`
+- `sqlfluff`
 
 These shims do not replace compilers. They find the real compiler later in PATH, execute it, preserve its output and exit code, then emit ANCP JSON.
 
@@ -125,4 +142,3 @@ A shim is production-acceptable when it obeys these rules:
 - it validates emitted ANCP documents.
 
 That gives agents a stable protocol without forcing every language compiler project to accept upstream patches first.
-

docs/repository-structure.md

@@ -89,7 +89,7 @@
 - `ancp`: the reference CLI for manifest, capability, check, explain, repair-plan, verify, graph, skills, validate, render, schema, and shim installation.
 - `ancp.proxy`: prefixed compiler proxies such as `ancp-cargo`, `ancp-tsc`, `ancp-python`, `ancp-kotlinc`, and `ancp-julia`.
 - `ancp.shim`: native-name wrappers installed by `ancp install-shims`, so normal commands such as `cargo check` and `python -m py_compile app.py` can emit ANCP sidecars.
-- `ancp.adapters`: native-tool adapters for Python, TypeScript, JavaScript, Rust, Go, C/C++, Java, Kotlin, C#/.NET, Swift, Zig, Ruby, PHP, Dart, Scala, and Julia.
+- `ancp.adapters`: native-tool adapters for Python, TypeScript, JavaScript, Rust, Go, C/C++, Java, Kotlin, C#/.NET, Swift, Zig, Ruby, PHP, Dart, Scala, Julia, Shell, PowerShell, Lua, Perl, R, Haskell, OCaml, Erlang, Elixir, Clojure, Nix, Terraform, Dockerfile, SQL, JSON, TOML, and YAML.
 - `ancp.resources`: packaged copies of the ANCP schema and taxonomies for installed environments.
 
 ## Contract Artifacts

docs/toolchains.md

@@ -0,0 +1,115 @@
+# Toolchain Coverage
+
+ANCP adapters are split into two layers:
+
+1. built-in protocol and parser support that ships with the Python package,
+2. native compiler/checker integrations that require the relevant ecosystem tool to be installed.
+
+The reference package is useful even when some compilers are missing. Missing tools are reported as ANCP `tool_failed` results instead of invalid JSON or fake success.
+
+## Check Local Availability
+
+Run:
+
+```bash
+python tools/check_toolchains.py
+```
+
+For machine-readable output:
+
+```bash
+python tools/check_toolchains.py --json
+```
+
+By default the checker returns exit code `0` so it can be used as a report in local verification. Add `--strict` when a CI job expects every tool in that job to be installed:
+
+```bash
+python tools/check_toolchains.py --language rust --language go --strict
+```
+
+## Built In
+
+These adapters do not require an external compiler for syntax validation:
+
+- JSON
+- TOML
+- YAML
+
+Python and PowerShell usually exist on Windows machines already, but they are still external runtime/toolchain integrations.
+
+## Recommended Native Tools
+
+| Language family | Tool |
+| --- | --- |
+| Python | `python` |
+| TypeScript | `tsc` |
+| JavaScript | `eslint` |
+| Rust | `cargo`, `rustc` |
+| Go | `go` |
+| C | `gcc` or `clang` |
+| C++ | `g++` or `clang++` |
+| Java | `javac` |
+| Kotlin | `kotlinc` |
+| C#/.NET | `dotnet` |
+| Swift | `swift` |
+| Zig | `zig` |
+| Ruby | `ruby` |
+| PHP | `php` |
+| Dart | `dart` |
+| Scala | `scala-cli` or `scalac` |
+| Julia | `julia` |
+| Shell | `shellcheck` preferred, `bash` fallback |
+| PowerShell | `pwsh` or Windows PowerShell |
+| Lua | `luac` preferred, `lua` fallback |
+| Perl | `perl` |
+| R | `Rscript` |
+| Haskell | `ghc` |
+| OCaml | `ocamlc` |
+| Erlang | `erlc` |
+| Elixir | `elixirc` |
+| Clojure | `clj-kondo` |
+| Nix | `nix-instantiate` |
+| Terraform | `terraform` |
+| Dockerfile | `hadolint` |
+| SQL | `sqlfluff` |
+
+## Windows User-Level Install Examples
+
+Use these only for tools you want to exercise locally. CI should install toolchains in dedicated jobs rather than relying on one giant image.
+
+```powershell
+scoop install shellcheck lua perl terraform hadolint
+npm install -g typescript eslint
+python -m pip install sqlfluff
+```
+
+Heavier ecosystems are better installed with their official installers or CI setup:
+
+- Rust: `rustup`
+- Java/Kotlin/Scala: JDK plus Kotlin/Scala tooling
+- Haskell: GHCup
+- OCaml: opam
+- Erlang/Elixir: official installers or package manager images
+- R: CRAN R installer
+- Swift, Zig, Dart, Ruby, PHP, Nix: ecosystem-specific installers
+
+## CI Strategy
+
+Do not require every compiler in one job. Use a matrix:
+
+- core job: schema validation, unit tests, package build, JSON/TOML/YAML/Python checks,
+- systems job: Rust, Go, C/C++, Zig,
+- JVM/.NET job: Java, Kotlin, Scala, C#,
+- dynamic job: JavaScript, TypeScript, Ruby, PHP, Dart, Julia, Lua, Perl, R,
+- functional job: Haskell, OCaml, Erlang, Elixir, Clojure,
+- infrastructure job: Shell, PowerShell, Nix, Terraform, Dockerfile, SQL.
+
+Each job should run:
+
+```bash
+python tools/check_toolchains.py --json --strict --language rust --language go
+python tools/run_bug_corpus.py
+ancp validate .ancp/bug-corpus
+```
+
+The matrix should fail when an expected installed compiler returns `tool_failed`, but the local all-language corpus may still pass schema validation when a developer has partial tooling.

docs/verification.md

@@ -78,6 +78,7 @@ ancp capabilities | python -m json.tool
 ancp schema | python -m json.tool
 python -m build
 python -m twine check dist/*
+python tools/check_toolchains.py
 ```
 
 Purpose:
@@ -87,6 +88,7 @@ Purpose:
 - confirms top-level CLI commands emit parseable JSON,
 - confirms wheel and source distributions can be built,
 - confirms package metadata is acceptable to Python packaging tools.
+- reports which native language toolchains are available for real compiler-backed smoke coverage.
 
 Expected result:
 
@@ -95,6 +97,7 @@ Expected result:
 - CLI JSON commands parse,
 - build succeeds,
 - `twine check` reports `PASSED` for every built artifact.
+- installed native toolchains appear in the availability report; missing optional toolchains are expected on partial developer machines.
 
 ## Pass 5: Invisible Compiler Layer And Bug Corpus
 

docs/vision-doctrine.md

@@ -57,7 +57,7 @@ The developer should not need to learn a new workflow just to make code agent-re
    If a compiler/tool is not installed or a language lacks stable structured output, ANCP must report that honestly as valid ANCP JSON. It must not fabricate successful checks.
 
 6. **Multi-language from the start.**
-   The important targets include Rust, Kotlin, Python, Julia, TypeScript, C, C++, Go, Java, C#/.NET, Swift, Zig, Ruby, PHP, Dart, Scala, and JavaScript.
+   The important targets include Rust, Kotlin, Python, Julia, TypeScript, C, C++, Go, Java, C#/.NET, Swift, Zig, Ruby, PHP, Dart, Scala, JavaScript, Shell, PowerShell, Lua, Perl, R, Haskell, OCaml, Erlang, Elixir, Clojure, Nix, Terraform, Dockerfile, SQL, JSON, TOML, and YAML.
 
 7. **Security is part of the protocol.**
    Build/test/repair commands can execute arbitrary code. ANCP must label effects, safety levels, and risky operations instead of hiding them.
@@ -271,4 +271,3 @@ The first public release succeeds if someone can:
 9. get a compact, useful diagnosis instead of a noisy compiler dump.
 
 That is the revolution: compilers become agent-readable without developers changing how they code.
-

examples/buggy/clojure/broken.clj

@@ -0,0 +1,4 @@
+(ns broken.core)
+
+(defn run []
+  (println "hello clojure")

examples/buggy/dockerfile/Dockerfile

@@ -0,0 +1,4 @@
+FROM ubuntu:latest
+RUN apt-get update && apt-get install -y curl
+COPY missing-file /app/
+CMD ["bash"]

examples/buggy/elixir/broken.ex

@@ -0,0 +1,5 @@
+defmodule Broken do
+  def run(name) do
+    IO.puts("hello #{name}"
+  end
+end

examples/buggy/erlang/broken.erl

@@ -0,0 +1,5 @@
+-module(broken).
+-export([main/0]).
+
+main() ->
+    io:format("broken erlang~n"

examples/buggy/haskell/Main.hs

@@ -0,0 +1,7 @@
+module Main where
+
+main :: IO ()
+main = do
+  putStrLn "broken haskell"
+  let x =
+  print x

examples/buggy/json/broken.json

@@ -0,0 +1,5 @@
+{
+  "name": "broken-json",
+  "enabled": true,
+  "items": [1, 2,]
+}

examples/buggy/lua/broken.lua

@@ -0,0 +1,4 @@
+local function broken(name)
+  print("hello " .. name)
+if true then
+  broken("lua")

examples/buggy/nix/broken.nix

@@ -0,0 +1,7 @@
+{ pkgs }:
+
+pkgs.mkShell {
+  packages = [
+    pkgs.git
+  ;
+}

examples/buggy/ocaml/main.ml

@@ -0,0 +1,4 @@
+let greet name =
+  print_endline ("hello " ^ name
+
+let () = greet "ocaml"

examples/buggy/perl/broken.pl

@@ -0,0 +1,7 @@
+use strict;
+use warnings;
+
+my $name = "perl";
+if ($name eq "perl" {
+  print "broken\n";
+}

examples/buggy/powershell/broken.ps1

@@ -0,0 +1,5 @@
+function Invoke-Broken {
+  param(
+    [string]$Name
+  Write-Output "Hello $Name"
+}

examples/buggy/r/broken.R

@@ -0,0 +1,7 @@
+broken <- function(x) {
+  if (x > 0) {
+    print("positive")
+  else {
+    print("negative")
+  }
+}

examples/buggy/shell/broken.sh

@@ -0,0 +1,5 @@
+#!/usr/bin/env bash
+if [ "$1" = "run" ]; then
+  echo "running"
+else
+  echo "missing fi"