Files
dotfiles/agents/.agents/skills/claude-subagent-playbook/references/prompts/triage-header.md
T
Dmitry Dementiev 10f55f35f2 fix(skills): решение координатору идёт с вариантами и ценой
- Зачем:
  - находка FIT-5 холодного ревью: развилка, сжатая до открытого вопроса,
    не даёт принять решение — кода нет в контексте ни у координатора, ни у
    владельца, а исполнитель в этот момент единственный держит картину.
- Что:
  - NEEDS_DECISION в triage-header.md требует варианты и рекомендацию
    исполнителя, а не только формулировку вопроса.
  - требование «понятно без чтения кода» адресовано обеим меткам
    координатора сразу — вместе с OUT_OF_SCOPE.
  - раздел «Пользователь в петле» SKILL.md обязывает координатора нести
    решение владельцу тем же составом.
- Проверка:
  - первым боевым прогоном конвейера: у NEEDS_DECISION в отчёте
    исполнителя есть варианты, их цена и рекомендация.
2026-08-04 21:44:49 +03:00

47 lines
2.3 KiB
Markdown

# Triage of findings from independent blind reviews of your implementation
Below are the findings of the blind reviews, **verbatim**. Process every
finding ID; change nothing outside their scope. Assign each ID exactly one
disposition:
- FIXED — a TDD fix: first a test reproducing the finding, then the fix;
name the check that proves closure;
- REJECTED — a reasoned rejection via contract/spec and test; a bare
"not important" is not accepted;
- OUT_OF_SCOPE — needs a boundary decision; state which one;
- NEEDS_DECISION — a product, architecture, or risk choice a technical
argument cannot settle; state the open question, the options you see,
and which one you would take and why. This includes the case where your
own fix would settle such a choice — the size of the edit is irrelevant.
OUT_OF_SCOPE and NEEDS_DECISION belong to the coordinator — do not
silently decide them yourself. Write both so they can be judged without
reading the code: name what each option costs and what it breaks. The
coordinator holds no code in context, by design.
## The review is evidence, not orders
A review report is a set of claims, not a list of instructions. Verify
every finding yourself. You own the task; the review is there to help.
- **Rejection is a normal outcome, not a conflict.** A REJECTED with an
argument from contract and test is worth exactly as much as a FIXED. A
run with no rejection and no finding closed by your own route says
nothing about the quality of the review — it is a reason to re-read
what you took on trust.
- **The reviewer's route is a proposal.** Closing a finding another way
is fine when the argument is stronger; name the route you took and why.
No commits. Do NOT edit the task file: write detailed resolutions and
rejection arguments to your assigned report file — re-checks and fresh
threads will read that file by path; the task file is updated separately
at the end of the run.
Final report: one line per ID — `ID | disposition | one-phrase essence`;
then disposition counters, test command with counters, list of changed
files, and open questions for the coordinator, if any.
Orchestrator notes, if present below, carry task-level decisions (out of
scope / risk accepted) and verified facts you don't have — they are not
findings to fix.