Muhammad Zakir Ramadhan
7fd578e13b
Approval was a list of tool names: `bash` needed it, `read_file` did not. That
fails in a specific way. `bash` covers `git status` and `rm -rf` equally, so a
user working through a batch presses `a` — always allow — on the first prompt and
every later command runs unasked, including the one they would have refused. The
gate was strongest when it mattered least and gone by the time it mattered.
Rules now match the *subject* of a call: the command for `bash`, the path for a
file tool, the pattern for a search.
"permission": {
"bash": { "*": "ask", "git *": "allow", "rm *": "deny" },
"edit_file": { "*": "deny", "src/generated/*": "allow" }
}
Plain last-match-wins, with no special case for deny. An earlier version made
deny win wherever it sat, on the theory that a refusal should be impossible to
undo by accident, and it made default-deny-with-exceptions unexpressible — which
is the shape a careful user actually writes, and the same shape as `*.env` denied
while `*.env.example` is allowed. Refusals that must never be configurable stay
in the guard plugin, which runs ahead of this and which --yolo cannot reach.
Three behaviours fall out of it:
- `always` grants the pattern the tool suggests, not the tool. Approving
`git status` runs `git log` unprompted and still asks about `npm publish`.
- `.env`, `.env.*`, and `.pem` are denied on read by default. Not gated, refused:
a secret that reaches the context is on the wire and in the session file, and
there is no taking it back. `.env.example` stays allowed.
- A call repeated identically three times in one turn asks even when allowed. A
model repeating itself is not making progress, and `bash: allow` is a statement
about which commands are safe rather than permission to loop.
The prompt now says which rule matched and what `always` would grant:
bash wants to run
git status --porcelain
y allow once | a always allow bash git * | n deny
A typo in a decision string is dropped at parse time, leaving the tool on its
default. Treating an unparseable value as `allow` would mean one misspelling
silently removing the gate.
Written after surveying Claude Code, Codex, opencode, and phi. Three of the four
had already moved to per-pattern rules; the credential deny and the repeat guard
come from opencode directly. ROADMAP records what was deliberately not taken and
why — OS sandboxing needs three platform implementations and is worse than
nothing if half-built, and opencode's own docs say LSP integration is often not a
net positive.
581 tests, up from 572. The engine is a pure function tested on its own, and the
loop is tested through a real Session: an allowed pattern never prompts, a denied
one never executes, and a `.env` read leaves no secret in the transcript.
Shiro Neko
An agentic coding CLI. It reads your code, edits it, runs your tests, and asks when the request is ambiguous — in a terminal UI, with every mutating action gated behind an approval prompt.
Install
A single prebuilt binary. No runtime, no node_modules.
# macOS, Linux
curl -fsSL https://raw.githubusercontent.com/zakirkun/shiro-neko/main/scripts/install.sh | sh
# Windows
irm https://raw.githubusercontent.com/zakirkun/shiro-neko/main/scripts/install.ps1 | iex
Both verify the download against the release checksums before installing. Builds are
published for linux-x64, linux-arm64, darwin-x64, darwin-arm64, and windows-x64.
Or from source:
git clone https://github.com/zakirkun/shiro-neko
cd shiro-neko
bun install
bun run install:local # builds and puts `shiro` on PATH
First run
shiro
With no API key configured it opens provider setup: pick an endpoint, paste a key, choose
from the models that endpoint actually reports. Settings land in
~/.shiro-neko/config.json. Run /provider any time to change them.
shiro-neko 0.1.0-beta.3 openai/gpt-5 session 0193ab2c
agent: default thinking: medium
cwd: /home/you/project
skills: debug, refactor, review, test
plugins: guard, time
approvals: ask for write_file, edit_file, multi_edit, bash, mcp__*
/help for commands
> why does the pagination test fail?
What it does
Answers about your code, grounded in your code. grep goes through ripgrep when it is
installed and honours .gitignore. list_dir gives an ignore-aware tree so it stops globbing
blindly to orient, and read_many_files pulls a batch in one round trip. read_file refuses
binaries rather than filling the context with mojibake.
Edits with your approval, gated per command. write_file, edit_file, multi_edit, and
bash stop for a y/a/n decision, with a coloured diff for edits. Rules match the command or
path rather than the tool, so git * can run unprompted while everything else still asks —
answering a whitelists that pattern, not the whole tool. .env and .pem files are refused on
read outright. The guard plugin refuses irreversible commands ahead of any of it — rm -rf,
git reset --hard, force pushes, DROP TABLE — and --yolo cannot bypass it.
Shows its work. Reasoning streams to a collapsed panel you can expand with ctrl-r, the
tool in flight is named as it runs, and bash output streams live instead of arriving all at
once when the command exits. ctrl-c kills a runaway command without ending the turn.
Takes prompts while it works. Type during a turn and it queues; the queue drains in order
when the turn ends. esc interrupts and clears it. @ completes workspace paths.
Reads git without touching it. git_status, git_diff, git_log, git_show, and
git_blame are approval-free, because they spawn git with a fixed argument list and cannot
mutate anything.
Asks instead of guessing. When a request has two readings that lead to different work, the agent puts a question on screen with options.
Delegates searches. task spawns a read-only subagent whose findings come back as one
message, so a search across forty files does not fill the main context. Its progress
streams to a panel.
Extensible from the prompt. /registry browses external skills and plugins and installs
them with one confirmation. A skill is shown in full before its text joins your system prompt;
a plugin is a manifest of refusal rules, never code.
Remembers between sessions. Decisions, working commands, and traps go into per-project memory that is injected at the start of every future session.
Survives long tasks. The task list and project memory live outside the message array,
so they survive both automatic pruning and /compact.
Runs headless. shiro -p "review this diff" --json for scripts and CI.
Keeps the tool list affordable. Fourteen built-in tools, grouped into sets. Each costs
about 550 characters of schema on every request, so { "toolSets": [] } trims back to the six
core ones and a disabled set reaches neither the wire nor the prompt.
Documentation
Start with whichever question you have. Each guide says what it decided and why, not just what the flags are.
| Guide | Contents |
|---|---|
| Configuration | config file, provider presets, environment, every flag |
| Tools | every tool, tool sets and what they cost, the approval model |
| Permissions | allow/ask/deny rules, patterns, defaults, the repeat guard |
| Agents and thinking | variants, thinking levels, step caps, which to reach for |
| Skills | the bundled skills, writing your own, why the catalogue is split |
| Plugins | the interface, the guard and its limits, builtin versus installed |
| Registry | installing external skills and plugins, publishing your own |
| Memory and state | memory, task lists, sessions, compaction and its repair |
| MCP | connecting servers, namespacing, cost, debugging one |
| Headless mode | -p, JSON events, exit codes, CI recipes |
| Architecture | how the loop works and why it is built this way |
| Development | building, testing, adding a tool, releasing |
| Roadmap | what is next and what has been declined |
| TODO | the current work list, with known rough edges |
Commands
Type / and a menu appears, narrowing as you type.
/help /agent [name] /think [level] /provider /models /model <id>
/skills /plugins /registry [search|add|remove] /init /context
/todos /notes /memory /tools /compact /cost
/sessions /resume <id> /save /clear /exit
esc dismisses a panel, interrupts a running turn, and clears the queue. ctrl-c kills the
running command but keeps the turn. ctrl-r expands the reasoning panel. @ completes a
workspace path. Up and down recall earlier prompts.
Status
Working: the agent loop, tool approvals, subagents, skills, plugins, per-project memory,
session persistence, MCP, markdown rendering, headless mode, five-platform builds, streaming
reasoning display, the mid-turn prompt queue, gateable tool sets, read-only git tools, batch
reads, @file completion, interruptible commands, and the external registry.
Next up is in TODO.md; the longer view and what has been declined are in
ROADMAP.md. The short version of what is missing: a summary of what compaction
discarded, web_fetch, a spend ceiling, and a cheaper model for subagent searches.
License
MIT. See LICENSE.