feat: zed target — .zed detection and Zed's silent frontmatter rules

[singhharsh1708]

The bug

Zed reads the vendor-neutral .agents/skills/<name>/SKILL.md path — Kitbash was already emitting it. But nothing in a Zed-only repo announces that: the agents adapter probes .agents/ and .codex/, and Zed creates neither.

So a repo whose only agent marker was .zed/ compiled to nothing but the eager AGENTS.md floor, paying a skill's whole body every session on an agent that can lazy-load it for free. For prereview that is ~560 standing tokens instead of ~40 — the 13× tax the benchmark exists to measure, charged to a target that never needed to pay it.

The target

zed detects .zed/ and emits the byte-identical file agents does, so a repo with both compiles that skill once, not twice. Ten targets, nine output files.

It also enforces two constraints Zed applies and KSF does not. Both fail silently in Zed — the skill is dropped at load with no diagnostic in the UI, which is the capability loss spec §2 forbids:

• name must match ^[a-z0-9]+(-[a-z0-9]+)*$. KSF's NAME_RE is looser and permits a doubled or trailing hyphen, so tidy--commits is a legal skill name Zed will not load. Zed's 64-character cap is unreachable from here — KSF caps names at 41.
• description must be non-empty and at most 1024 UTF-8 bytes, not characters — a trap for non-ASCII text, where 600 characters can be 1,200 bytes. Manifested skills are bounded at 200 characters (600 bytes at worst), so only an unmanifested SKILL.md, whose frontmatter is copied through unvalidated, can breach either bound.

⚠ tidy--commits → zed: name must match ^[a-z0-9]+(-[a-z0-9]+)*$ — no leading,
  trailing or doubled hyphen; Zed drops the skill at load with no diagnostic
⚠ wide-desc → zed: description is 1200 bytes, over Zed's 1024-byte limit
  (the limit is bytes, not characters); Zed drops the skill at load with no diagnostic

Both are warnings, so compile --strict fails on them.

No disable-model-invocation key is emitted. Zed accepts one, but KSF has no field meaning "never auto-invoke" — triggers.auto is a hint for auto-invocation — so synthesizing a value would guess at the author's intent and could suppress a skill they expect to fire.

The honesty bug this surfaced

kitbash explain <skill> zed answered "no capability degradation" for a skill Zed discards entirely. Literally true, reads as false, and it is exactly the class of silent loss this project exists to prevent — in our own tool.

Adapters can now declare target-specific constraints through an optional lint() separate from emit(). explain reads it directly rather than emitting files to find out.

Verification

• 9 new e2e assertions: .zed/-only detection, frontmatter shape, no false positives on a conforming skill, shared-path single-write with no conflict, each of the three constraint breaches, --strict failure, and explain surfacing the rejection.
• npm run check clean · full suite green · npm run bench deterministic · node site/build.mjs --check current.
• Benchmark regenerated with the zed row; docs, site, config id table and target counts updated (9 → 10 targets; "nine files" prose left alone, since it is still nine files).

Release

Version bumped to 0.12.0 with a CHANGELOG section. Tag v0.12.0 after merge to trigger release.yml.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
kitbash	Ready	Preview, Comment	Jul 28, 2026 12:37pm