Jinka

Drop Markdown wish files, run one command, and let Claude Code carry them out on your repositories.

Source repo
yegor256/jinka
Stars
★ 20
Last updated
3d ago
License
MIT
Primary language
JavaScript

At a glance

How it runs
CLI
Works with
Platform-specificClaude Code
Cost
Free software; you pay for model usage
Setup effort
Low · running in minutes
You'll need
Node.js (npm global install)Claude CodeShell / CLINetwork accessLocal filesystem
Typical use
Developers already using Claude Code who want reusable Markdown instructions instead of retyping prompts.
Not a fit if
  • Teams that don't want a Claude Code dependency
  • Users who prefer a GUI over the terminal
  • Organizations requiring switchable model providers
Maintenance
Commits in the last 30 dayslast commit 2026-11-18latest release 0.0.14 (2026-11-17)

What does this agent do, and when should you use it?

Jinka is a Node.js command-line orchestrator distributed as the npm package @yegor256/jinka under the MIT license. You describe what you want in a Markdown file placed under ~/.jinka/wishes/ — the README example asks for a GitHub repository, foo/bar, to be kept in good shape — and then invoke the jinka command, which reads those wishes and leans on Claude Code to fulfill them. The repository bills itself as a "Simple Command Line Orchestrator of Agents" and carries topics such as coding-agent, harness and llm-agents. There is no hosted service or UI: the execution boundary is your local terminal plus the local filesystem, and the docs stop at the quickstart level.

Jinka treats Markdown files in ~/.jinka/wishes/ as its input; each file holds a natural-language wish such as "I wish my GitHub repository to be in good shape: foo/bar." Running jinka hands those wishes to Claude Code, which performs the actual work against the named target, for example inspecting or tidying a GitHub repository. The end-to-end loop is: author a wish file, install globally via npm, run jinka, and the wish is acted upon. The README does not specify output format, log locations, exit codes, or retry behavior.

  1. Developers already using Claude Code who want reusable Markdown instructions instead of retyping prompts.
  2. Maintainers who want to trigger a repository health check or cleanup from the terminal with one plain-language sentence.
  3. Solo developers who prefer file-as-configuration over learning a complex CLI flag set.
  4. Engineers who want to keep several routine wishes side by side in ~/.jinka/wishes/ and run them in bulk.
  5. Teams evaluating a minimal agent orchestrator before building custom scripting around an LLM.

Key setup facts

Pulled from the install and usage notes below. The project README has the current values.

Install / run
  • npm install -g @yegor256/jinka
Config files
~/.jinka/wishes/Make-Me-Happy.md

How do you install or deploy this agent?

You need Node.js with npm, plus a working Claude Code setup.

npm install -g @yegor256/jinka

After installation the jinka command is available in your shell.

How do you use this agent?

Create a wish file first:

~/.jinka/wishes/Make-Me-Happy.md

With content like:

I wish my GitHub repository to be in good shape: `foo/bar`.

Then run it:

jinka

The README does not document flags, whether you pick a wish interactively, or whether every file in the directory runs at once.

What are this agent's strengths and limitations?

Pros
  • Wish-as-file interface: describe the goal in Markdown instead of memorizing CLI flags.
  • Very short install path — one global npm command gets you running.
  • Built on Claude Code, so it reuses an existing coding agent rather than shipping its own LLM layer.
  • MIT licensed, so you can fork and modify freely.
Limitations
  • Core operation depends on Claude Code; the README documents no alternative model or provider path.
  • Documentation is a single quickstart: no output format, error handling, logging or concurrency details.
  • No UI or hosted option — terminal only.
  • Only one example wish, so task coverage and best practices are unclear.

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

Agent Source review Form / cost Stars Updated Language Full support on
Jinka This agent 41 · Major gaps CLIFree + model costs ★ 20 3d ago JavaScript Claude Code
Ordewell 87 · Good CLIFree + model costs ★ 211 1d ago TypeScript Codex · Claude Code · OpenAI API · Claude API
Cersei — Rust SDK for Coding Agents 56 · Major gaps Library / SDKFree + model costs ★ 467 3mo ago Rust OpenAI API · Claude API
Nori Skillsets 45 · Major gaps CLIFree ★ 170 1d ago TypeScript Codex · Claude Code

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Major gaps
41/ 100 5-point scale 2.1 / 5
Trust 10/29
Reliability 6/14
Adaptability 10/18
Convention 6/18
Effectiveness 6/13
Verifiability 3/8
Why each dimension lost points
Trust10 / 29 · 1.7/5

The skills impose explicit gates before external side effects (approving review, passing checks, no conflicts) and forbid subagents and out-of-scope actions, which supports least_privilege and external_effects; however merging PRs and filing issues are irreversible external actions with no human confirmation step, so user_confirmation scores 1. Neither README nor CLAUDE.md states which model or service receives wish contents, code, or logs, so data_flow_transparency and sensitive_data_handling score 1 on the strength of the 'treat PR/logs as data' instruction alone. Dependencies are only commander, cross-spawn, slice-ansi with no lockfile or audit note, so dependency_security scores 1. Rollback is limited to an archive directory and append-only triaged.csv; there is no mechanism to undo a merged PR or a filed issue, so rollback scores 1. LICENSE and SPDX headers exist but publisher identity is unverified, so source_attribution scores 1.

Reliability6 / 14 · 2.1/5

The skill texts are internally consistent: each declares it will not build, test, or dispatch subagents, and this does not conflict with the CLAUDE.md README constraint, so self_consistency scores 2. Dependency availability: engines requires Node>=20.12, dependencies use floating ^ ranges with no lockfile, so dependency_availability scores 1. Failure messages: test assertions carry readable messages, but user-facing runtime error text is not visible in the sample, so failure_messages scores 1.

Adaptability10 / 18 · 2.8/5

Skill descriptions give clear trigger phrases ('merge PR', 'check CI', 'find a new bug') and stop conditions, so trigger_precision and capability_boundaries each score 2; audience_and_scenarios covers repo maintenance, CI triage, and backlog cleanup, scoring 2. Environment fit mentions only GitHub, the gh CLI, and Node version, with no guidance for non-GitHub platforms or missing permissions, so environment_fit scores 1.

Convention6 / 18 · 1.7/5

README is deliberately minimal and CLAUDE.md forbids usage details there, leaving information architecture thin, so information_architecture scores 1; install steps (npm install -g) are clear, so install_notes scores 2. Version is 0.0.0, so naming_stability scores 1. There is one wish example and no FAQ, so examples_and_faq scores 1. No known-limitations section exists, so known_limitations scores 0. LICENSE is a complete MIT text, so license scores 2. No CHANGELOG, so versioning_changelog scores 0. Maintenance responsibility appears only via the package.json author field and the publisher is unverified, so maintenance_responsibility scores 1.

Effectiveness6 / 13 · 2.3/5

Skill output conventions (title format, snippet quoting, terse wording) make results directly usable on GitHub, so output_usability scores 2. Value depends on external Claude Code and gh credentials, and the marginal value of auto-merging and auto-filing issues is only moderate without evaluation evidence, so marginal_value scores 1; each run costs model calls and GitHub API traffic with no cost-control guidance, so cost_benefit scores 1.

Verifiability3 / 8 · 1.9/5

Skill gates (approving review, check status) trace to GitHub API queries, so claim_traceability scores 1; test files and source modules (agent, appended, archived) corroborate each other, so cross_source_corroboration scores 1; however the README's 'your wish will come true' promise is not separated from what the skills actually do, so fact_inference_separation scores 1.

Risks and how to mitigate them
  • Skills auto-merge PRs and file issues without human confirmation, which are irreversible external side effects; restrict repository scope and review the gates before enabling.
  • Neither README nor CLAUDE.md states which model or service receives wish contents, code, and logs, leaving the data flow opaque.
  • Dependencies use floating ranges with no lockfile, so installs may drift over time.
  • Version is 0.0.0, so interfaces and behavior may be unstable; publisher identity is unverified.
Evidence confidence: Low Reviewed Nov 18, 2026 Reviewed revision ce0fb2f235d2
See the full review method →
Maintain this project? Add the FARS badge to your README

The badge shows the current review score and verdict and links back to this page. It refreshes about once an hour after a re-review.

FARS

[![FARS review](https://followagents.com/badge/jinka.svg)](https://followagents.com/en/agents/jinka)
<a href="https://followagents.com/en/agents/jinka"><img src="https://followagents.com/badge/jinka.svg" alt="FARS review"></a>

Recent releases

Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.

  1. 0.0.14
    Release notes excerpt
    What's Changed * Show the summary of a JSON answer in the thread row by @yegor256 in https://github.com/yegor256/jinka/pull/227 * chore(deps): update reviewdog/action-actionlint action to v1.84.1 by @renovate[bot] in https://github.com/yegor256/jinka/pull/229 * chore(deps): update dependency eslint to v10.15.0 by @renovate[bot] in https://github.com/yegor256/jinka/pull/230 * chore(deps): update crate-ci/typos action to v1.52.2 by @renovate[bot] in https://github.com/yegor256/jinka/pull/228 …
    View on GitHub ↗
  2. 0.0.13
    Release notes excerpt
    What's Changed * chore(deps): update dependency slice-ansi to v10.0.1 by @renovate[bot] in https://github.com/yegor256/jinka/pull/219 * Paint the token count red when it is over one million by @yegor256 in https://github.com/yegor256/jinka/pull/222 * Run every wish in its own temporary directory under ~/.jinka by @yegor256 in https://github.com/yegor256/jinka/pull/223 * Repair red or conflicting pull requests opened by this account by @yegor256 in https://github.com/yegor256/jinka/pull/225 Full Changelog: https://github.com/yegor256/jinka/compare/0.0.12...0.0.13
    View on GitHub ↗
  3. 0.0.12
    Release notes excerpt
    What's Changed * Forbid the agent to schedule work for later by @yegor256 in https://github.com/yegor256/jinka/pull/180 * Forbid the agent to write a wish similar to an existing one by @yegor256 in https://github.com/yegor256/jinka/pull/182 * Make every skill line direct and consistent with the prompt by @yegor256 in https://github.com/yegor256/jinka/pull/185 * Ask the agent for a JSON answer with summary and link by @yegor256 in https://github.com/yegor256/jinka/pull/186 * Rest a thread between wishes with a visible countdown by @yegor256 in https://github.com/yegor256/jinka/pull/188 …
    View on GitHub ↗

FAQ

Do I need an API key beyond installing jinka?
The README mentions no key configuration. Because it works through Claude Code, credential needs follow your Claude Code setup, which the README does not detail.
Must wish files live in ~/.jinka/wishes/?
The README only demonstrates Make-Me-Happy.md at that path and does not say whether a custom directory or a file argument is supported.
What kinds of targets or tasks can it handle?
The sole example is a maintenance wish for one GitHub repository (foo/bar); supported task types and permission scope are not listed.
What happens when a wish fails?
No failure reporting, retry or rollback behavior is documented, so you need to watch the run yourself.
Is Jinka free?
The software is free, but Jinka calls an AI model, and you pay that model provider for usage (or run a local model).
Is Jinka open source?
Yes. The source code is on GitHub (yegor256/jinka) under the MIT licence.
Is Jinka safe to use?
FollowAgents' FARS review (Nov 18, 2026) scored Jinka 10/29 on trust and safety. None of the eight safety controls the review checks for was missing outright.
Is Jinka still maintained?
Yes. The latest commit on GitHub is from 2026-11-18, within the last 30 days. The latest release is 0.0.14 (2026-11-17).
Who makes Jinka?
Jinka is published on GitHub by the account yegor256. FollowAgents has not verified which person or company is behind that account.
Was this page helpful?
View on GitHub ↗ Install ↓

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related concepts