Use when the user asks to create, write, compose, or design a new agent skill — turning an idea, notes, requirement, or repeated workflow into a SKILL.md with eval cases and a trigger set through an interview → draft → critique → refine loop.
SkillScore73Maintained
Automatic
◆AI Engineer pack
Last verified 8/4/2026
What it changes
Passed 15 of 21 test cases with the skill, and 11 of 21 without it.
Without skill50%
With skill73%
Checks it passed that the model alone missed
Directs running shakespii lint with the --json flag.
Specifies using shakespii init to scaffold the skill.
States that trigger classification accuracy must reach at least 0.8.
and 4 more in the full report
7 fixed9 passed either way4 still failing2 broke
Pass-rate lift
+23 pp
Tokens
+21%
Agent turns
0%
Response time
-55%
Model gemini-3.6-flashMeasured 8/4/2026v116 of 22 cases passed with skill · 21 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 7 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 7
Prompt
What trigger classification accuracy score must the skill achieve on the harness before trigger refinement can be considered complete?
What was checked, without and with the skill
Checked
Without skill
With skill
States that trigger classification accuracy must reach at least 0.8.
Missed
Passed
case-20 · gemini-3.6-flash · 8/4/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md144 lines3 files
authoring-skills
Intent
Turn a human's idea into a finished Agent Skill through a structured loop:
interview the human for the raw material, draft against the anatomy contract,
critique with a rubric of qualities no linter can check, and refine until the
harness — not taste — says the skill works. The using-shakespii skill teaches
how to drive the CLI; this skill decides what the new skill should say.
Inputs
The idea: a problem statement, requirement, or repeated workflow the human
wants captured as a skill.
A writable parent directory for the new skill.
Optional: raw material the human already has — notes, transcripts, a real
worked example, memory excerpts.
Preconditions
The shakespii CLI resolves (shakespii --version succeeds); setup lives in
the using-shakespii skill's Preconditions.
The using-shakespii skill is available — every CLI mechanic here (fix loop,
eval runs, trigger measurement) delegates to it.
A human is reachable for the interview, or the task prompt already supplies
and approves the interview's answers.
Procedure
Phase 1 — Interview. Ask one question at a time, multiple-choice where the
options are enumerable, until every anatomy section has raw material:
Intent: what problem, for whom, and what does a successful use look like?
Triggers: at least five real requests that should fire the skill, and at
least three lookalikes that must not.
Inputs and preconditions: what the skill consumes; binaries, paths, and
environment it assumes.
Procedure: walk one real occurrence of the workflow end to end.
Example: one real input with its real output — not an invented pair.
Failure modes: what has gone wrong when this was done by hand.