What does this skill do, and when should you use it?
This skill is a behavioral convention document inside the manaflow-ai/cmux repository, aimed at developers or coding agents working on cmux, a Ghostty-based macOS terminal. It sets three rules: when a behavior is exposed through multiple surfaces (keyboard shortcut, command palette, context menu, CLI/socket command, settings UI, debug menu), implement one shared action/model path; optimistic updates must keep a single mutation path with request-id or snapshot tracking and explicit rollback on failure; and when a user reports tests missed a bug, behavior-level coverage around the exact repro path must be added before the fix is claimed complete. It contains no scripts — it is instruction text that an agent reads and follows.
- Enumerates the surfaces a behavior may span (keyboard shortcut, command palette, context menu, CLI/socket, settings UI, debug menu) and mandates one shared action/model path
- Forbids patching a single surface while others keep duplicated logic
- Specifies the optimistic-update convention: record pending state with a request id or previous snapshot, reconcile from the authoritative result, and roll back or surface an error state on failure
- When tests missed a bug, requires behavior-level coverage on the exact repro path before declaring the fix done
- An agent adding a new entrypoint to an existing cmux feature (e.g., a command-palette entry for a settings action) that must reuse the shared path instead of copying logic
- Fixing a bug in optimistic UI updates where a defined mutation path and rollback convention are needed
- Responding to a user report that tests missed a bug, by adding behavior-level coverage on the repro path first
- Reviewing a change to check whether it patched one surface but left the others with stale duplicated logic
- Not applicable outside the cmux codebase — the entrypoint list (debug menu, CLI/socket commands, etc.) is written around cmux's own structure
- Not for end users or people who merely run the cmux terminal; it targets contributors and agents modifying cmux's code
How do you install this skill?
- This is a static source review; nothing was executed and confidence is low.
- The skill is a convention document with no scripts or tests; its effect depends on whether the agent follows the instructions and cannot be statically verified.
- File-level license metadata is NOASSERTION; the applicable license is the repo-level GPL-3.0-or-later (with BUSL-1.1 for web/ and related directories).
- The publisher is not verified by the FollowSkills registry; identity is unknown.
- Content is English-only with no Chinese support; value is bound to the cmux (macOS) ecosystem and is largely inapplicable to other repositories.
- No version number or changelog; instructions may drift as the repository evolves.
- Local filesystem
cmux (the macOS terminal app from manaflow-ai/cmux)
The skill lives at skills/cmux-shared-behavior/ in the manaflow-ai/cmux repo (one of 25 bundled skills). The source documents no dedicated install command for this skill — it ships with the repository; clone the repo and place the skill folder into your agent's skills directory.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/manaflow-ai/cmux.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/cmux-shared-behavior" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- I'm adding a command-palette entry to cmux's notification-clearing action — what does the cmux-shared-behavior skill require?
- A user says our tests missed a split-pane resize bug; add behavior-level coverage on the exact repro path per the shared behavior rules before we call it fixed
- This setting's optimistic update is duplicated across the keyboard shortcut and CLI paths — refactor it to one shared mutation path
The skill triggers via its description: whenever a task involves a behavior exposed through multiple entrypoints, optimistic updates, or a test that missed a bug, the agent loads it and follows the rules. Typical flow: before changing a cmux behavior, check which surfaces expose it, implement a single shared action/model path, and verify each entrypoint; for optimistic updates, follow the single-mutation-path convention; for reported missed bugs, add behavior-level tests on the exact repro path first. The skill ships no runnable scripts or commands.
What are this skill's strengths and limitations?
- Rules are concrete and actionable: named surface list, a specified mutation-path pattern, and a defined missed-bug workflow
- Lightweight plain text any file-reading agent can follow, with no runtime dependencies
- Derived from real maintenance conventions of the cmux codebase, addressing a genuine class of problem there (duplicated multi-surface logic)
- Only relevant to the cmux codebase; little transferable value for other projects
- Convention-only — no accompanying scripts or automated enforcement
- Repository license is flagged NOASSERTION on GitHub; the skill file's exact licensing scope needs separate verification
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| cmux Shared Behavior Rules this page | 48 · Use with care | ★ 28k | 1d ago | NOASSERTION |
| cmux Workspace Skill | 64 · Recommended | ★ 28k | 1d ago | NOASSERTION |
| cmux Diagnostics | 51 · Use with care | ★ 28k | 1d ago | NOASSERTION |
| cmux Testing Skill | 60 · Recommended | ★ 28k | 1d ago | NOASSERTION |
| cmux Settings Management Skill (cmux-settings) | 58 · Recommended | ★ 28k | 1d ago | NOASSERTION |
Unlike generic code-style or testing-convention skills, this is an architecture-level behavior convention for a single app (cmux); cmux's GUI-plus-CLI surface model differs from tmux-style pure CLI multiplexers, and this rule exists to keep those surfaces consistent.
How did FollowSkills review this skill?
The skill is pure guidance: it requests no permissions, runs no scripts, touches no network, and handles no sensitive data, so least-privilege is trivially satisfied and there is no red-line risk. However, data-flow transparency and rollback are normative prose rather than verifiable mechanisms; source attribution rests on repo-level licensing (GPL-3.0-or-later, file metadata NOASSERTION) and an unverified publisher, so full marks are withheld.
The instructions are self-consistent and non-contradictory; the three rules (shared entrypoints, single optimistic-update path with rollback, missed-bug coverage) are clearly stated. But as an abstract convention with no scripts, examples, edge-case handling or failure-feedback design, static review can only confirm readability, not key-path reproduction; heavily deducted.
The trigger description is fairly precise (behaviors exposed through multiple entrypoints, optimistic UI/CLI flows, missed-bug scenarios) and the audience is cmux contributors/agents. But no non-fit boundaries, no input/output examples, no Chinese-language support, and value is bound to the cmux/macOS ecosystem, limiting general applicability.
Name and description match the content and the structure is short and clear; the repo has clearly layered GPL-3.0 + BUSL licensing and active CI maintenance signals. But the skill itself has no version, changelog, FAQ, examples, or known-limitation disclosure, and file-level license metadata is NOASSERTION.
As a coding-convention skill, its value — injecting multi-entrypoint consistency and optimistic-update rollback rules into agent behavior — is plausible and non-trivial. But static review cannot verify it actually improves outputs, and there are no representative outputs to audit; deducted.
The skill content is author-asserted engineering convention with no tests or example outputs that could be independently reproduced. The repository has real CI and test suites, but none covering this skill's own key paths; only thin primary material (SKILL.md, agents/openai.yaml) is auditable.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →