How do I install the Skillassay skill?

Run npx skillmds@latest add rakib-nyc/skillassay in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Skillassay safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Skillassay?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Skillassay free to use?

Yes. Installing skills from SkillMD is free. This skill is licensed under Apache-2.

Who published Skillassay?

rakib-nyc (@rakib-nyc) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Skillassay by rakib-nyc — Agent Skill

Name: Skillassay
Author: rakib-nyc

Skillassay

Audits AI coding-agent context. Measures always-on context-window cost of CLAUDE.md, AGENTS.md and Agent Skills, and reports skills that will not load, stale path references, redundant instructions and duplicate skill names. Use when the user asks to audit, measure or reduce context cost, check whether a skill is valid or will load, review CLAUDE.md or AGENTS.md quality, or find duplicate, conflicting or unused skills.

rakib-nyc 1cda9f2 2.5 KB Updated Sep 17, 2026

File contents

skillassay

Run the assay CLI and interpret its report. All analysis happens in the CLI. Never estimate a token count, invent a percentage, or judge a finding yourself — report what the tool returns.

Commands

npx skillassay .                      # whole repository
npx skillassay path/to/SKILL.md       # one skill — use this when authoring
npx skillassay . --json               # structured output, nothing truncated
npx skillassay . --top 10             # bounded human-readable output
npx skillassay . --fix                # unified diff of deletions; never applied
npx skillassay --explain SPEC-NAME-INVALID   # a rule, its source, its limits

Deciding what to do first

Read conformance.willLoad from --json before anything else.

false — one or more skills violate the Agent Skills specification and no compliant client will register them. Fix every SPEC-* error first; findings about verbosity or duplication are moot until then.
true — the skills load. Now the warn findings are worth acting on.

Reading the report

ALWAYS-ON CONTEXT — tokens loaded every session before the user types. Counted with cl100k_base as a proxy, which the output labels; do not present it as an exact Claude count.
Found but not counted — real files this harness does not load. Expected, not an error.
FINDINGS — each carries a rule ID, a citation and, where applicable, the tokens saved by acting on it.

Rules for you

Show the user the --fix diff before applying any part of it. The tool never writes to files and neither should you without agreement.
When a finding is info, say so. RED-STALE-PATH in particular cannot tell a stale reference from a directory created at runtime.
If the user asks why a rule exists, run --explain <RULE-ID> rather than guessing; it prints the source and what the rule does not establish.
Skills meant to work across Claude Code, Codex and Gemini CLI belong in .agents/skills/, which every compliant client reads.

---name: skillassaydescription: Audits AI coding-agent context. Measures always-on context-window cost of CLAUDE.md, AGENTS.md and Agent Skills, and reports skills that will not load, stale path references, redundant instructions and duplicate skill names. Use when the user asks to audit, measure or reduce context cost, check whether a skill is valid or will load, review CLAUDE.md or AGENTS.md quality, or find duplicate, conflicting or unused skills.license: Apache-2.0---# skillassayRun the `assay` CLI and interpret its report. **All analysis happens in theCLI.** Never estimate a token count, invent a percentage, or judge a findingyourself — report what the tool returns.## Commands```bashnpx skillassay . # whole repositorynpx skillassay path/to/SKILL.md # one skill — use this when authoringnpx skillassay . --json # structured output, nothing truncatednpx skillassay . --top 10 # bounded human-readable outputnpx skillassay . --fix # unified diff of deletions; never appliednpx skillassay --explain SPEC-NAME-INVALID # a rule, its source, its limits```## Deciding what to do firstRead `conformance.willLoad` from `--json` before anything else.- **`false`** — one or more skills violate the Agent Skills specification and no compliant client will register them. Fix every `SPEC-*` error first; findings about verbosity or duplication are moot until then.- **`true`** — the skills load. Now the `warn` findings are worth acting on.## Reading the report- **ALWAYS-ON CONTEXT** — tokens loaded every session before the user types. Counted with cl100k_base as a proxy, which the output labels; do not present it as an exact Claude count.- **Found but not counted** — real files this harness does not load. Expected, not an error.- **FINDINGS** — each carries a rule ID, a citation and, where applicable, the tokens saved by acting on it.## Rules for you- Show the user the `--fix` diff before applying any part of it. The tool never writes to files and neither should you without agreement.- When a finding is `info`, say so. `RED-STALE-PATH` in particular cannot tell a stale reference from a directory created at runtime.- If the user asks why a rule exists, run `--explain <RULE-ID>` rather than guessing; it prints the source and what the rule does not establish.- Skills meant to work across Claude Code, Codex and Gemini CLI belong in `.agents/skills/`, which every compliant client reads.

Frequently asked questions