Platform Reports

Table of Contents

For AI agents: a documentation index is available at /llms.txt — markdown versions of all pages are available by appending index.md to any URL path.

We ran the same 38 automated checks against each platform and compared what actually happened. Each row below asks one question about platform behavior; the cells say in plain language what each platform did. Rows marked 📌 are where platforms disagree: the cases where a skill that works on one platform behaves differently on another.

Full detail for every finding (the exact verdict, how content reached the model, confidence, and notes) lives on the per-platform pages. Each page opens with a spec alignment summary: where that platform’s observed behavior contradicts or matches what the Agent Skills specification prescribes, and how it handles skills that violate the spec’s format rules.

For what these findings mean when writing a skill, see the cross-platform authoring guidance. The check list has the full rationale behind every question. All findings come from headless sessions, which can differ from interactive use; outcomes marked † rest on behavioral inference rather than direct transcript evidence. Two checks (context compaction protection, trust gating) need interactive sessions and are marked manual.

Loading Timing #

When skill content enters the model’s context, and how much loads at each stage.

Question	Antigravity CLI	Claude Code	Codex CLI
Does the harness read only SKILL.md metadata at discovery, or the full body?	Metadata only †	Metadata only	Metadata only
On activation, does the harness load only the SKILL.md body, or also bundled resources, and by which vehicle?	Body only	Body only	Body only
Does activation pre-fetch files markdown-linked from the SKILL.md body, and does that extend to a file mentioned only as plain text?	No pre-fetching †	No pre-fetching	No pre-fetching

Directory Recognition #

Which directories a platform treats as part of a skill, and what happens to ones it doesn’t recognize.

Question	Antigravity CLI	Claude Code	Codex CLI
Are the three spec directories (scripts/, references/, assets/) enumerated to the model at activation?	Nothing enumerated †	Nothing enumerated	Nothing enumerated
📌 Is a resources/ directory (alternative to spec’s references/) loaded, enumerated, readable, or invisible?	Readable when the model looks	Not surfaced; model never looked	Not surfaced; model never looked
What happens to directories the spec never named (evals/, templates/): injected, readable on demand, or invisible?	Not surfaced; model never looked †	Not surfaced; model never looked	Not surfaced; model never looked

Resource Access Patterns #

How supporting files (scripts, references, assets) become available to the model.

Question	Antigravity CLI	Claude Code	Codex CLI
At activation, are a skill’s reference files enumerated to the model (names), loaded outright (contents), or invisible until explored?	Nothing enumerated †	Nothing enumerated	Nothing enumerated
📌 When the model follows a SKILL.md relative path like references/setup-guide.md, what does it resolve against, and does the bare path work as written?	Bare path fails; model recovers	Bare path fails; model recovers	Model used full paths (base untested) †
With two skills both owning references/API.md, does the activated skill’s read get its own file or the sibling’s?	Got its own file †	Got its own file †	Got its own file †
Can the model read outside the activated skill’s directory (a sibling skill’s file), and is anything visibly blocked?	Reads outside the skill allowed	Reads outside the skill allowed	Reads outside the skill allowed
How deep in the directory tree do reference files stay reachable? Rungs at one, two, three, and five levels.	All depths reachable (tested to 5)	All depths reachable (tested to 5)	All depths reachable (tested to 5)
📌 Can the agent run a bundled scripts/ file and receive its output?	Script ran; output returned	Blocked with an error	Script ran; output returned

Content Presentation #

What the model actually sees, at discovery and at activation, and how it’s formatted.

Question	Antigravity CLI	Claude Code	Codex CLI
📌 Which frontmatter fields does the discovery listing surface to the model: name and description only, or also compatibility, metadata values, or file locations?	Also surfaces location †	Name and description only	Also surfaces location
📌 Does the SKILL.md YAML frontmatter reach the model at activation, or only the body?	Visible (model reads the raw file)	Stripped before injection	Visible (model reads the raw file)
📌 Is injected skill content wrapped in structured tags, or delivered as raw markdown, and what does the model see on pull harnesses?	Raw file via model read	Raw markdown, no wrapper tags	Raw file via model read

Lifecycle Management #

How skill content is managed over the course of a conversation.

Question	Antigravity CLI	Claude Code	Codex CLI
📌 When the same skill is activated twice in one session, is its content loaded again or deduplicated?	Model re-reads each time	Full content re-injected every time	Model re-reads each time
After SKILL.md is edited mid-session, does reactivation serve the fresh content or a cached copy?	Edits picked up immediately	Edits picked up immediately	Edits picked up immediately
Is skill content protected when the context window fills up?	manual	manual	manual

Access Control #

How platforms gate skill loading and handle control-related frontmatter.

Question	Antigravity CLI	Claude Code	Codex CLI
Do project-level skills require trust approval before loading?	manual	manual	manual
Does a compatibility field naming another platform gate loading, get surfaced to the model, or get ignored?	Loads normally, no gating	Loads normally, no gating	Loads normally, no gating
Does the experimental allowed-tools field pre-approve anything, compared against an identical skill without it?	Ran with and without the field	Ran with and without the field	Ran with and without the field

Skill-to-Skill Invocation #

Whether one skill’s instructions can activate another skill.

Question	Antigravity CLI	Claude Code	Codex CLI
Can one skill’s instructions get a second installed skill activated by name?	Second skill activated	Second skill activated	Second skill activated
How deep does a skill→skill→skill activation chain run before something stops it?	Full three-skill chain completed	Full three-skill chain completed	Full three-skill chain completed
When two skills each instruct activating the other, does the A→B→A cycle loop, get blocked, or stop by model choice?	Model stopped the loop itself †	Model stopped the loop itself	Model stopped the loop itself
Does the invoke chain still complete when the activation prompt is Japanese?	Full three-skill chain completed	Full three-skill chain completed	Full three-skill chain completed

Skill Dependencies #

What happens when one skill depends on another, formally or in prose.

Question	Antigravity CLI	Claude Code	Codex CLI
Is a dependency expressed only in prose (“now activate the invoke-beta skill”) resolved at runtime?	Second skill activated	Second skill activated	Second skill activated
📌 When a skill instructs activating a skill that is not installed, is the failure visible, reported, or silently skipped?	Reported missing without attempting	Failed with a visible error	Reported missing without attempting
Does the platform act on nonstandard dependency frontmatter (requires, depends-on, priority)?	Ignored †	Ignored	Ignored
📌 Can a project-level skill invoke a dependency that exists only at user level, and what is the failure mode when it is absent?	Resolved across scopes; missing dependency reported, not attempted	Resolved across scopes; missing dependency fails visibly	Resolved across scopes; missing dependency reported, not attempted

Discovery Scope #

Where platforms look for skills: which directories are scanned, how deep, and what happens when two discovered skills claim the same name.

Question	Antigravity CLI	Claude Code	Codex CLI
📌 Is a skill installed only at the cross-client .agents/skills convention path discovered?	Convention path is the native directory †	Convention path not scanned	Convention path scanned
📌 Does the skills root get scanned recursively (a skill under a grouping directory), and is a SKILL.md outside any root discovered?	Direct children only; stray file ignored †	Direct children only; stray file ignored	Scans the root recursively; stray file ignored
📌 Is a SKILL.md nested inside another skill’s references/ tree discovered as a separate skill?	Not discovered †	Not discovered	Discovered as a separate skill
📌 With the same skill name installed at project and user scope, which variant’s content activates?	Project scope wins †	User scope wins	Lists both; model chose the project variant †

Validation Strictness #

How strictly platforms judge skills that break the spec’s format rules: rejected, repaired, or loaded anyway.

Question	Antigravity CLI	Claude Code	Codex CLI
📌 Is a skill whose description holds an unquoted colon (invalid YAML) still discovered and loadable?	Not cataloged; file still readable †	Tolerated and loaded	Tolerated and loaded
📌 Is a skill with no description field skipped (as the guide prescribes), or loaded anyway?	Loaded anyway †	Loaded anyway	Skipped (as the guide prescribes)
📌 Are skills whose names break the spec’s rules (uppercase, consecutive hyphens, over 64 characters) still discovered and loadable?	All three invalid names tolerated †	All three invalid names tolerated	Tolerated only: uppercase, double-hyphen
📌 When directory name and frontmatter name disagree, which identity is the skill listed and invocable under?	Frontmatter name wins †	Directory name wins	Listed under both names
Is a skill whose metadata frontmatter holds nulls and empty strings still discovered and loaded, and do those keys reach the model?	Loaded fine	Loaded fine	Loaded fine
Is a skill whose description exceeds the spec’s 1024-character limit still discovered, and does the full value survive untruncated?	Loaded anyway †	Loaded anyway	Loaded anyway
Is a skill whose compatibility value exceeds the spec’s 500-character limit still discovered and loadable?	Loaded anyway †	Loaded anyway	Loaded anyway