AGENT SKILL

performance-awareness

How the performance score moves for your role. Use when the score context shows a declining trend, or before handing off work you have not self-checked against every criterion.

License 许可
Apache-2.0
最近更新
2026-11-15
Stars
★ 122
Forks
18

01 前言 Front Matter

name
performance-awareness
category
quality
description
How the performance score moves for your role. Use when the score context shows a declining trend, or before handing off work you have not self-checked against every criterion.

03 SKILL.md 正文

Performance Awareness

Overview

Every agent carries a performance score (0–100, starts at 100). It is not decoration: it reflects how much rework your output causes downstream, and your run context already injects it (agent.score_context) — this skill explains what moves it. A task that sails through its gates clean protects your score; one that bounces back at any gate costs you. The score rewards getting it right the first time.

Core principle: The cheapest revision is the one that never happens. Every gate you pass clean is points kept; every bounce is points lost AND time lost.

How the score moves

Event Delta Who is charged
Your task reaches done or released +5 each the assignee
Revision requested from code_review −10 the in_progress owner
Revision requested from ready_for_qa or in_qa −10 the in_progress owner
Human UAT fails −8 the in_progress and in_qa owners
A human overturns your approval at the same review gate (review escape) −10 the reviewer who approved it
QA: a bug it caught +6 QA, per bug
QA: a scenario it confirmed valid / invalid +1 / −1 QA, per scenario
QA: finished testing a task +5 QA

The asymmetry is deliberate: one revision (−10) wipes out two clean releases (+10), and a defect that escapes every earlier gate and fails Human UAT (−8) still costs real points.

The gates your work passes

in_progress → code_review → ready_for_qa → in_qa → human_uat → done → released
              (architect)                  (QA)    (human)

A defect caught at code_review costs the developer −10. The SAME defect that slips to in_qa or human_uat still costs the developer and now also charges whoever owned the column it escaped through, plus the time of every role between. Catch your own defects before handoff — that is what the score is measuring.

To protect your score

  1. Read ALL acceptance criteria before starting. Most bounces are unmet AC, not bugs.
  2. Self-check every AC against fresh, executed evidence before moving the task forward or recording a verdict.
  3. Never guess a product decision — add_task_comment with numbered questions instead. A wrong guess becomes a UAT failure, and UAT failures charge more than one role.
  4. On a revision, address EVERY point and fix at the root, not the symptom. A partial fix bounces again for another −10.
  5. Run the real checks in THIS run before claiming anything passes or is complete.

Worked Example

Two developers each ship 5 tasks in a week.

  • Dev A rushes: 5 tasks to code_review fast, 3 bounce for unmet AC (−30), fixes them, all 5 eventually released (+50: +25 done, +25 released). Net for the week: +20, plus the architect reviewed 8 times.
  • Dev B self-checks every AC before handoff: 5 tasks, 0 bounces, all released (+50). Net: +50, architect reviewed 5 times.

Same 5 tasks shipped. The difference is entirely self-verification before handoff.

Red Flags

  • Moving a task forward "to see if it passes" — the next gate is not your test suite.
  • Skipping an AC, or a scenario, because "it's probably fine."
  • Leaving a revision point partially addressed.
  • Approving something you did not actually verify — a review escape costs as much as a bounced revision, and it is charged to you alone.

04 同仓库更多 Skills

makifbaysal
makifbaysal/tasktrooper

visual-craft

暂无简介

★ 122
makifbaysal
makifbaysal/tasktrooper

ux-copy

Use when you write any words a user reads in a mockup or a hand-off - content language and voice, buttons, labels, errors, empty states, confirmations, numbers and dates, and the longest-string check

★ 122
makifbaysal
makifbaysal/tasktrooper

screen-mockup-html

Use when you draw screens as the design task's HTML mockup - ONE canvas document holding every screen and every variant side by side, the sanitizer-safe recipe, phone and wide frames, the states grid, static control states, and the loopback self-check with screenshots at 375, 768 and 1440 attached to the task

★ 122
makifbaysal
makifbaysal/tasktrooper

design-variants

Use when a screen design needs variants - how many, variants as side-by-side rows on the one canvas, what makes two variants genuinely different, recommending one, how the human's choice is read, and narrowing to the chosen one in revision

★ 122
makifbaysal
makifbaysal/tasktrooper

design-system-authoring

Use when you derive, change or extend a design system - where tokens live per stack, the DTCG token tree, role-based names for both themes, base vs layer vs drift, the DESIGN.md section order, the inventory format, the contrast check and the lint every proposal returns

★ 122
makifbaysal
makifbaysal/tasktrooper

design-review

Use when a design task asks you to compare a built screen with its approved design - screenshots at 1440, 768 and 375 attached to the task, Blocker/High/Medium/Nitpick triage, evidence per finding, problems not prescriptions

★ 122