| name | effort-aware |
| description | Demonstrates branching skill behavior on the current effort level using ${CLAUDE_EFFORT} (Claude Code v2.1.120+). Activates when discussing how to scope work across effort levels. |
| allowed-tools | Read, Grep, Glob |
Effort-Aware Skill
Demonstrates how a skill can adapt its instructions to the current effort
level. Claude Code injects ${CLAUDE_EFFORT} into skill text at load
time (v2.1.120+), so the same skill can give terse advice on low and
exhaustive advice on xhigh without needing separate skills.
Current effort: ${CLAUDE_EFFORT}
Guidance for this effort level
If ${CLAUDE_EFFORT} is low or medium:
- Make the smallest correct change. Don't refactor surrounding code.
- Skip exploring alternatives — pick the obvious path.
- One pass, no self-review beyond a re-read.
- Don't write tests for trivial changes (one-liner config, doc fix).
If ${CLAUDE_EFFORT} is high:
- Sketch 2–3 approaches before picking one. Note the tradeoffs in your
response.
- Add tests for new behavior.
- Self-review the diff before reporting done.
If ${CLAUDE_EFFORT} is xhigh or max:
- Map invariants the change must preserve. Read enough surrounding code
to be confident.
- Build a sequence of small, reversible commits if the change is large.
- Tests cover happy path, error path, and at least two boundary cases.
- Run a
pr-reviewer self-review pass before reporting done.
- Note follow-ups (debt, deferred work) explicitly so the human can
triage.
Why this exists
Older Claude Code versions had no way for skills to adapt their tone to
the user's effort setting. You either had to write conservatively (and
under-deliver on xhigh) or aggressively (and over-engineer on low).
Effort interpolation closes that gap.
When this skill activates
When the user discusses scoping, planning a change, or asks how
thoroughly to do something. The skill helps you calibrate effort to
match ${CLAUDE_EFFORT} rather than picking a single fixed default.
See also
