Supported Harnesses
This page is the canonical 5-tier support matrix for AI coding harnesses ("agents") that Spec Kitty integrates with. Each row conforms to the HarnessEntry schema in kitty-specs/spec-kitty-3-2-docs-01KS4KSZ/data-model.md. The research procedure that backs each classification lives in docs/development/3-2-harness-research-method.md.
Access date for citations: 2026-06-03.
Tier legend
| Tier | One-line definition |
|---|---|
| first_class | Installed by default; canonical UX (slash command or Agent Skill); integration-tested; current external doc citation. |
| supported | Installed by the standard installer; current external doc citation; may lack first-class integration tests or have a degraded mechanism. |
| partial | Installer touches the host but coverage is incomplete, or no current external citation could be located. |
| experimental | Provisional coverage that may break without notice; typically depends on another harness's installation. |
| archived | No longer covered by the installer, or the upstream project is deprecated; coverage (if any) lives in docs/migration/. |
Full tier criteria and promotion rules are maintained in docs/development/3-2-harness-research-method.md §5 and §6.
Support matrix
| Harness | Key | Installed surface | Mechanism | Tier | Citation (accessed 2026-06-03) | Notes |
|---|---|---|---|---|---|---|
| Claude Code | claude |
.claude/commands/ |
slash_command | first_class | https://docs.claude.com/en/docs/claude-code/overview | Reference harness; integration-tested in src/specify_cli/. |
| Codex CLI | codex |
.agents/skills/spec-kitty.*/SKILL.md |
skill | first_class | https://github.com/openai/codex | Heaviest Agent Skills integration; canonical skill tree at .agents/skills/. |
| OpenCode | opencode |
.opencode/command/ |
slash_command | supported | https://opencode.ai/docs | /spec-kitty.* command set installed via standard installer. |
| Cursor | cursor |
.cursor/commands/ |
slash_command | supported | https://cursor.com/docs | /spec-kitty.* command set installed via standard installer. |
| Google Gemini CLI | gemini |
.gemini/commands/ |
slash_command | supported | https://github.com/google-gemini/gemini-cli | /spec-kitty.* command set installed via standard installer. |
| Qwen Code | qwen |
.qwen/commands/ |
slash_command | supported | https://github.com/QwenLM/qwen-code | /spec-kitty.* command set installed via standard installer. |
| Amazon Q CLI | q |
.amazonq/prompts/ |
prompt | supported | https://docs.aws.amazon.com/amazonq/latest/qdeveloper-ug/ | Retained as legacy alongside Kiro rebrand per CLAUDE.md. |
| GitHub Copilot | copilot |
.github/prompts/ |
prompt | supported | https://docs.github.com/en/copilot/customizing-copilot/adding-custom-instructions-for-github-copilot | Prompt-file mechanism; /spec-kitty.* set installed. |
| Augment Code (Auggie) | augment |
.augment/commands/ |
slash_command | supported | https://docs.augmentcode.com/auggie/overview | /spec-kitty.* command set installed via standard installer. |
| Roo Cline | roo |
.roo/commands/ |
slash_command | supported | https://docs.roocode.com/ | /spec-kitty.* command set installed via standard installer. |
| Kilo Code | kilocode |
.kilocode/workflows/ |
workflow | supported | https://kilocode.ai/docs | Workflow mechanism; /spec-kitty.* set installed. |
| Windsurf | windsurf |
.windsurf/workflows/ |
workflow | supported | https://docs.windsurf.com/windsurf/cascade/workflows | Workflow mechanism; /spec-kitty.* set installed. |
| Kiro | kiro |
.kiro/prompts/ |
prompt | partial | https://kiro.dev/docs | Prompt-file surface only for Spec Kitty 3.3 plugin-install scoping; not a plugin/Powers bundle candidate until upstream exposes and verifies a package primitive. |
| Google Antigravity | antigravity |
.agent/workflows/ |
workflow | partial | https://www.antigravity.google/docs/rules-workflows | Local live CLI verification was unavailable on 2026-06-03; installed surface remains documented as partial until current Antigravity CLI workflow paths are smoke-tested. |
| Pi TUI | pi |
.agents/skills/spec-kitty.*/SKILL.md |
skill | partial | https://pi.dev/docs/latest/skills | Installer-supported command-skill packages; keep partial until a real-session smoke test is recorded. |
| Mistral Vibe | vibe |
.agents/skills/ via .vibe/config.toml skill_paths |
skill | experimental | https://docs.mistral.ai/mistral-vibe/terminal/quickstart | Uses the shared .agents/skills/spec-kitty.* packages plus a .vibe/config.toml skill_paths entry. Behaviour may change with Vibe upstream. |
| Letta Code | letta |
.agents/skills/spec-kitty.*/SKILL.md |
skill | partial | https://docs.letta.com/letta-code/skills | Installer-supported command-skill packages; keep partial until a real-session smoke test is recorded. |
Cross-reference quick view
| Tier | Count | Members |
|---|---|---|
| first_class | 2 | Claude Code, Codex CLI |
| supported | 10 | OpenCode, Cursor, Gemini CLI, Qwen Code, Amazon Q CLI, GitHub Copilot, Augment Code, Roo Cline, Kilo Code, Windsurf |
| partial | 4 | Kiro, Google Antigravity, Pi TUI, Letta Code |
| experimental | 1 | Mistral Vibe |
| archived | 0 | |
| Total | 17 |
Promotion path
A harness moves up tiers only when new evidence lands. See docs/development/3-2-harness-research-method.md §6 for the full rule. Summary:
partial→supportedwhen (a) the installer produces a full/spec-kitty.*set, (b) a current external citation is recorded, and (c) at least one smoke test has been documented.supported→first_classwhen (a) integration tests exercise the harness end-to-end, (b) the mechanism is the harness's canonical UX, (c) the per-harness page underdocs/how-to/harnesses/<key>.mdis non-stub, and (d) the promotion is logged in CHANGELOG.- Demotion happens when citations cannot be re-verified across two freshness audits, the upstream is deprecated, or the installer ceases to produce a valid surface.
Tier moves are recorded in this matrix's row notes and the project CHANGELOG.
Maintenance
- The inventory step is re-run on every release; the citation step is re-run on every release as part of the freshness audit (WP13).
- This matrix is the published artifact; the procedure that backs it is maintained at
docs/development/3-2-harness-research-method.md. - Per-harness "how-to" pages live at
docs/how-to/harnesses/<key>.mdand are promoted under the plan's matrix-first default (decision01KS4KTS4V300M9MMTS1AJEGXY).