feat(agent): add windsurf as an install target

[Davidson3556]

Fixes #28

Describe the changes you have made in this PR -

Adds Windsurf (Codeium's Cascade editor) as an agent install target, so testsprite agent install --target windsurf and testsprite setup --agent windsurf wire the verification-loop skill into a Windsurf project.

• New own-file target windsurf → .windsurf/rules/testsprite-verify.md.
• Cascade frontmatter trigger: model_decision + description: — the equivalent of the Cursor .mdc alwaysApply: false mode (only the description is surfaced up front; Cascade pulls in the full rule body when the description shows it is relevant), which matches how this skill should activate.
• Budget handling: a .windsurf/rules/*.md file is capped at ~12 K characters and Cascade silently truncates beyond it. The full skill body (~21 KB) would be cut in half, so the windsurf target renders the compact body — the same trimmed variant the Codex/AGENTS.md target already ships (~5 KB) — selected via a new compactBody flag on the target spec. The rendered file is ~5.5 KB: well within budget, and it still carries the H1, the when-to-run guidance, and the load-bearing testsprite test run … --wait / test artifact get commands. agent.ts and renderForTarget choose the same body so the installed bytes equal the asserted render.
• Everything else derives from the TARGETS map automatically (the agent list table, the setup --agent choices, skill-nudge install detection). Updated the two hardcoded help strings in agent.ts, the --help snapshot, the unit tests (render test for the Cascade frontmatter + a budget test asserting the windsurf render stays under 12 K and uses the compact body), the e2e matrix guards / multi-target / content-integrity coverage, and the README / DOCUMENTATION target lists.

Demo/Screenshot for feature changes and bug fixes -

Tests:

npm test          

npm run test:e2e  # e2e  

npm run typecheck && npm run lint && npm run format:check

---

Code Understanding and AI Usage

Did you use AI assistance (ChatGPT, Claude, Copilot, etc.) to write any part of this code?

• No, I wrote all the code myself
• Yes, I used AI assistance (continue below)

If you used AI assistance:

• I have reviewed every single line of the AI-generated code
• I can explain the purpose and logic of each function/component I added
• I have tested edge cases and understand how the code handles them
• I have modified the AI output to follow this project's coding standards and conventions

Explain your implementation approach:

Problem: TestSprite's value is wiring the verify skill into coding agents, but Windsurf — a top-tier AI editor — wasn't a supported target.

The target system is a clean table (TARGETS in agent-targets.ts) with a per-target wrap() for the file's frontmatter, so the core addition is one entry plus a wrap function. I chose trigger: model_decision after checking Windsurf's rules docs: its semantics match the existing Cursor target (alwaysApply: false) — surface the description, load the body on demand — rather than always_on (injects the whole skill into every prompt) or manual (needs an @-mention, defeating an auto-verify skill).

The non-obvious part is the size budget. Windsurf caps a .windsurf/rules/*.md file at ~12 K characters and truncates silently beyond it; the full skill body renders to ~22 KB, so half the skill would be lost. The repo already maintains a trimmed/compact body for the Codex/AGENTS.md target, which is agent-agnostic (no Codex- or AGENTS.md-specific wording), so I reused it for Windsurf via a compactBody flag rather than authoring a third asset. Both the renderForTarget default and the agent.ts install path branch on this flag (loading each source body at most once), so what gets written equals what the unit test renders.

Alternatives considered: (1) reuse wrapMdc (Cursor's wrapper) — rejected, Cursor's description/alwaysApply keys are not Windsurf's trigger/description; (2) ship the full body and just emit a budget warning like the Codex target — rejected, because unlike Codex (whose body fits its 32 K budget) the full body always exceeds Windsurf's cap, so it would warn on every install and still truncate; (3) the legacy single-file .windsurfrules — rejected in favour of the current .windsurf/rules/*.md directory format.

Edge cases handled/tested: a render test asserts the Windsurf output carries trigger: model_decision + description: and not the name:/alwaysApply: keys; a budget test asserts the real windsurf render is < 12 K characters, uses the compact body (a full-body-only sentence is absent) yet keeps the load-bearing commands, and is smaller than the full claude render; the "install all own-file targets" unit test derives its list from OWN_FILE_TARGETS; and the e2e content-integrity loop gates the full-body assertion on non-compact targets while adding a Windsurf frontmatter branch.

---

Checklist before requesting a review

• I have added proper PR title and linked to the issue
• I have performed a self-review of my code
• I can explain the purpose of every function, class, and logic block I added
• I understand why my changes work and have tested them thoroughly
• I have considered potential edge cases and how my code handles them
• If it is a core feature, I have added thorough tests
• My code follows the project's style guidelines and conventions

---

Note: Please check Allow edits from maintainers if you would like us to assist in the PR.