[Feature]: Allow agent-namespaced branches (`claude/**`, `copilot/**`, etc.) to bypass numeric prefix requirement

[pradeepmouli]

Background

AI coding agents create branches autonomously when working on tasks. Each major agent has an established convention for its branch names — typically a short agent identifier followed by a slash and a description:

Agent	Example branch
GitHub Copilot	copilot/fix-login-null-check
Claude Code	claude/add-user-profile-page
Cursor	cursor/refactor-auth-module
Windsurf	windsurf/update-api-client
Gemini CLI	gemini/improve-test-coverage
Codex	codex/migrate-to-typescript
Trae	trae/fix-rendering-bug
Qwen Code	qwen/add-search-endpoint
opencode	opencode/cleanup-dependencies
VS Code (generic)	vscode/implement-dark-mode

These names are meaningful and consistent — the agent prefix makes it immediately clear which tool created the branch — but they don't match spec-kit's required ###-description pattern and are currently rejected by check_feature_branch().

Problem

When an AI agent creates a branch and then invokes a spec-kit command (e.g. specify tasks, specify implement), validation fails because the branch name doesn't follow the numeric prefix convention. The agent either has to abandon its natural naming scheme or the workflow silently breaks.

This is becoming more common as teams use AI agents as first-class contributors. Spec-kit is increasingly the workflow layer inside an agentic loop, not just a tool used by humans at a terminal.

Proposal

Recognise agent-namespace branches as inherently valid — no numeric prefix required for branches that match <agent>/**.

The simplest implementation is a built-in allowlist in check_feature_branch():

local agent_prefixes=(
    "claude/"
    "codex/"
    "copilot/"
    "cursor/"
    "gemini/"
    "opencode/"
    "qwen/"
    "trae/"
    "vscode/"
    "windsurf/"
)

for prefix in "${agent_prefixes[@]}"; do
    [[ "$branch" == "$prefix"* ]] && return 0
done

A more extensible alternative — consistent with the extension manifest direction — would be a top-level agent_prefixes key in extension.yml or a dedicated config, allowing new agents to be registered without a spec-kit core change:

# .specify/config.yml or extension.yml
agent_branches:
  prefixes:
    - "claude/"
    - "copilot/"
    - "cursor/"
    # ...

Either approach solves the immediate problem; the config-driven approach ages better as the agent landscape continues to evolve.

Why Not Just Use ###- Branches

Agents can be instructed to use ###- naming, but this:

• Requires per-agent prompt customisation to override their default behaviour
• Loses the agent provenance signal in the branch name (useful for auditing and PR review)
• Breaks when agents are updated and revert to their defaults
• Doesn't help in environments where the branch is created by the agent before spec-kit commands are invoked

Acceptance Criteria

• Branches matching any built-in agent prefix (claude/, copilot/, cursor/, gemini/, codex/, opencode/, qwen/, trae/, vscode/, windsurf/) pass check_feature_branch() without a numeric prefix
• The allowlist is either configurable or clearly documented so teams can add prefixes for internal or emerging agents
• specify doctor reports the active agent prefix configuration
• The specify tasks, specify implement, and related commands work correctly on agent-prefixed branches (spec path resolution, branch detection, etc.)

[mbachorik]

I do not like it - the index based spec (and branches names by specs) are great to stack changes when launching multiple agents and experiments using spec-kit. The AI coding agents are tools used by spec-kit (so they need to conform), not vice versa.

[mnriem]

I would favor --no-git here and then let the agent or yourself just create the branch