ShellSyntaxTree

A focused .NET library that parses bash and PowerShell command strings into a structured AST. Purpose-built for tools that need to reason about shell commands without running them — approval gates for LLM-emitted commands, CI/CD script auditors, sandbox policy generators, audit-log analytics.

Hand-rolled, AOT-trim friendly, zero native dependencies. Multi-targets netstandard2.0 and net8.0.

dotnet add package ShellSyntaxTree --version 0.2.0

The 0.2.0 release adds PowerShell support. The latest stable 0.2.x package supports bash and PowerShell together. The public surface documented below tracks the dev branch.

What you get

For an input like cd /repo && rm /etc/passwd, ShellSyntaxTree produces:

flowchart TD
    classDef bad fill:#fee,stroke:#b00,stroke-width:2px
    A[cd /repo<br/>📁 /repo] -- "&&" --> B[rm<br/>📁 /etc/passwd<br/>cwd: /repo]
    class B bad

Loading

A two-clause AST where the second clause's Args includes a synthetic /repo attribution arg (so consumers can see "this rm is implicitly operating in /repo") and /etc/passwd is resolved and marked IsPath = true. Hard-deny rules over /etc/* fire immediately; no substring matching, no shelling out, no false positives.

Things you can do with it

• Approval gates for AI agents — given a command emitted by an LLM, decide ALLOW / PROMPT / DENY before invoking the shell.
• CI/CD pipeline audits — scan shell steps in GitHub Actions / Jenkinsfile / Azure Pipelines for writes outside the workspace, curl | bash from non-allowlisted hosts, hardcoded credential echoes.
• Sandbox / container policy — derive the minimum-viable volume mount set or AppArmor profile from a build script.
• Pre-commit linters — flag dangerous patterns (rm -rf /, chmod 777 /etc/*) in shell scripts at commit time.
• Shell history / audit-log analytics — ingest ~/.bash_history or auditd records into structured form for SIEM-style insights.
• Documentation / explainers — convert complex one-liners into readable structure for tutorials and runbooks.

The original consumer is Netclaw's approval policy; the library is built to be reusable beyond that.

Quick start

using ShellSyntaxTree;

var parser = new BashParser();
var parsed = parser.Parse("cd /repo && rm /etc/passwd");

if (parsed.IsUnparseable)
{
    // Safe-fail: prompt the user, deny the command, etc.
    Console.WriteLine($"can't model: {parsed.UnparseableReason}");
    return;
}

foreach (var clause in parsed.Clauses)
{
    Console.WriteLine($"{clause.Operator} {clause.Verb.Joined}");

    foreach (var arg in clause.Args.Where(a => a.IsPath))
    {
        var marker = arg.IsCwdAttribution ? "↳ cwd" : "  path";
        Console.WriteLine($"    {marker}: {arg.Resolved}");
    }

    foreach (var redirect in clause.Redirects.Where(r => !r.IsDynamicSkip))
    {
        Console.WriteLine($"    {redirect.Direction}: {redirect.Target}");
    }
}

Run that against the example input and you get:

None cd
      path: /repo
AndIf rm
    ↳ cwd: /repo
      path: /etc/passwd

Consumer guide

The consumer guide develops the quick start into a production-oriented algorithm: parser selection, safe-fail handling, command identity, paths and cwd attribution, redirects, compounds and pipelines, and PowerShell-specific alias and dynamic-command behavior. It also links to immutable examples from Netclaw's live approval-gate integration.

Public API surface

namespace ShellSyntaxTree;

public interface IShellParser { ParsedCommand Parse(string command); }
public sealed class BashParser : IShellParser { /* … */ }
public sealed class PwshParser : IShellParser { /* … */ }   // v0.2.0

public abstract record ShellParserOptions { /* HomeDirectory, WorkingDirectory */ }
public sealed record BashParserOptions : ShellParserOptions;
public sealed record PwshParserOptions : ShellParserOptions;

public sealed record ParsedCommand { /* Source, Clauses, IsUnparseable, … */ }
public sealed record Clause        { /* Operator, Verb, Args, Redirects, Elements, IsSubshell, IsCommandStringWrapped */ }
public sealed record ClauseElement { /* Raw, Value, Role, source span, verb-relative position, path facts */ }
public sealed record VerbChain     { /* Tokens, Joined, CanonicalVerb, IsDynamic */ }
public sealed record Arg           { /* Raw, Resolved, Kind, IsPath, IsCwdAttribution, IsFlag */ }
public sealed record Redirect      { /* Direction, Target, IsDynamicSkip */ }

public enum ArgKind            { Literal, EnvVar, Glob, Tilde, DynamicSkip }
public enum ClauseElementRole  { Verb, Argument, Redirect }
public enum RedirectDirection  { In, Out, Append, ErrOut, ErrAppend }
public enum CompoundOperator   { None, AndIf, OrIf, Sequence, Pipe }

Both parsers emit the same ParsedCommand AST — a consumer walks a PowerShell parse exactly as it walks a bash one. A Windows cmd parser remains deferred.

Behavioral contract: SPEC.md (bash + shared surface) and SPEC.POWERSHELL.md (PowerShell).

Samples

Two runnable samples live under samples/.

ShellSyntaxTree.Cli.Sample — terminal explainer + audit policy

dotnet run --project samples/ShellSyntaxTree.Cli.Sample -- explain "cd /repo && rm /etc/passwd"
dotnet run --project samples/ShellSyntaxTree.Cli.Sample -- audit "cd /repo && rm /etc/passwd"

# --shell pwsh routes the same explain / audit logic through PwshParser:
dotnet run --project samples/ShellSyntaxTree.Cli.Sample -- explain --shell pwsh "gci C:\logs | rm"

explain pretty-prints the AST with [flag] / [path] / [cwd-attr] / [dyn-skip] / [glob] markers per arg. audit runs a small built-in policy ("deny writes in /etc, /usr, /bin, /sbin, /lib", "warn on curl | bash", "warn on dynamic args in path slots") and exits 0 / 1 / 2 by severity. See samples/ShellSyntaxTree.Cli.Sample/Commands/AuditPolicy.cs for the illustrative policy code.

ShellSyntaxTree.Web.Sample — Blazor WebAssembly Mermaid visualizer

Paste a bash or PowerShell script, watch the parsed AST render as a Mermaid flowchart in your browser. Everything runs client-side — pasted scripts never leave your machine. Useful for "what does this script actually do?" moments and for understanding how the library models constructs like subshells and bash -c / pwsh -Command recursion. Static Invoke-Expression / iex payloads surface the same way, while computed payloads safe-fail.

dotnet run --project samples/ShellSyntaxTree.Web.Sample
# → http://localhost:5239

A shell selector switches between the bash and PowerShell parsers. The presets demonstrate compound commands, Bash subshell isolation, PowerShell grouping and alias resolution, command-string recursion, dynamic-cwd attribution, and unparseable inputs. Each preset shows what the library produces in a single click.

Building from source

dotnet tool restore
dotnet build -c Release
dotnet test  -c Release
dotnet pack  -c Release -o ./bin/nuget

global.json pins the SDK; you need .NET 10 SDK or later for the .slnx solution format.

Versioning

Tags are bare SemVer version numbers — no v prefix. The release workflow asserts this and fails fast on misformatted tags.

• 0.1.x — bash parser. Additive after 0.1.0 (more verb table entries, more corpus, bug fixes).
• 0.2.0 — first PowerShell parser (PwshParser). Adds the shared ShellParserOptions base and the additive VerbChain.CanonicalVerb / VerbChain.IsDynamic fields; renames Clause.IsBashCWrapped → IsCommandStringWrapped (breaking — see RELEASE_NOTES.md).
• 1.0.0 — when an external consumer beyond Netclaw ships against it without finding API gaps.

License

Apache-2.0. Copyright © 2026 Aaron Stannard.

---

Repository layout — for contributors and curious agents:

Path	What
src/ShellSyntaxTree/	The library (bash + PowerShell parsers)
tests/ShellSyntaxTree.Tests/	xUnit unit tests + corpus runner
tests/ShellSyntaxTree.Tests/Corpus/<shell>/*.json	Corpus entries — the acceptance contract (bash + powershell)
samples/ShellSyntaxTree.Cli.Sample/	Console explainer + audit policy
samples/ShellSyntaxTree.Web.Sample/	Blazor WASM Mermaid visualizer
docs/CONSUMER_GUIDE.md	Production-oriented consumer algorithm + Netclaw case study
tools/PwshCorpusTool/	PowerShell corpus authoring aid
SPEC.md, SPEC.POWERSHELL.md	The behavioral contract
openspec/	Change-proposal history (rationale for design decisions)
PROJECT_CONTEXT.md, TOOLING.md, AGENTS.md	Repo governance — for autonomous agents
IMPLEMENTATION_PLAN.md	NOW / NEXT / LATER work tracker