CLI overview
The standalone asterweave command-line interface (preview): what Phase A does, how to run it from the repository, and the full command reference.
Authentication
How the Asterweave CLI signs in through gh and claude, what it stores, every reason code with its fix, and how to troubleshoot sign-in.
Configuration and trust
How the Asterweave CLI layers configuration, which keys exist, where its user files live, and how repository trust is granted, revoked, and invalidated.
Exit codes and doctor
Every exit code of the Asterweave CLI, the JSON error envelope, which stream gets what, and the stable doctor check ids and statuses.
Runs
Asterweave CLI runs (Phase B): run show, approve and abort, the delivery and bugfix graphs, budgets, approvals, crash recovery and the run lock. run start and run resume drive Claude Code; the development-only fake executor simulates everything.
Policy and guards
The Asterweave CLI guard: reason codes and decisions, supervised versus autonomous mode, policy.json, the allow-list shell classifier and its limits, the user-owned allowCommands list, how it differs from the plugin guard, scoped approvals, and the experimental hook pre-tool-use command.
Hooks
Asterweave CLI hooks: the events, the entry schema, user hooks versus trusted repository hooks, ordering, timeouts, failure policy, the stdin payload and the security model.
Tools, usage and routing
Asterweave CLI Phase C: the typed tool registry, the read-only GitHub adapter, the MCP stdio client and its approvals, token usage and budgets, and model routing. Fixture-verified only: no live model calls.
Sessions and worktrees
Asterweave CLI Phase D-a: durable sessions (start, list, search, resume, fork, archive, clean, transcript), schema migration and recovery, and isolated git worktrees. No runner yet: nothing here calls a model.
The Claude runner
asterweave run start and run resume with the Claude runner (Phase D-b): what runs, how it is guarded, the environment allow-list, failure classes and exit codes, sessions and worktrees, and what is still unverified.
Pull request and pipeline monitoring
Asterweave CLI Phase E: poll a pull request and its workflow runs, decide when a task is closable, and produce an explicit diagnosis packet. Polling only, read-only, zero model calls. Everything is fixture-verified; live GitHub is unverified.
HRIS-014 fixture demo
A self-contained, zero-dependency demo of the Asterweave CLI on one fictional task (HRIS-014): sign-in to a closable pull request, plus the failure paths. Everything is a fixture: no live model, no live GitHub, no live MCP server.
Image attachments
Attach PNG, JPEG and WebP images to a session or a run: accepted formats, the limits (guesses), what is stripped, exactly what is sent and when, trust, and what is still unverified.
Context and usage views
The asterweave context command and the interactive /context, /usage, /attachments and /image views: three separate numbers (context size, cumulative usage, budget remaining), measured versus unknown, and what nothing here estimates.
Plugins and MCP management
Asterweave CLI Phase F2: scoped, delegated management of Claude Code plugins (and MCP servers). Asterweave owns trust, approval and the effective view; claude performs the writes. Fixture-verified only.
Conversation memory
Asterweave CLI persistent conversation memory: a local SQLite store of conversation events, checkpoints and reviewed decisions, fed by asterweave run and injected into its model requests as a delimited historical-data section, with search, trace, retention, deletion and export commands. Unencrypted.