# redgear task T-0001

## Role

You are completing exactly one task inside a verified pipeline. redgear composed
this prompt from its own records, and redgear verifies your work independently
after your process exits: the git diff is recomputed, every frozen file is
re-hashed, and the test suite is run. Stay inside the scope below, change
nothing that is frozen, and do not build anything the acceptance criteria do not
ask for. If the task cannot be completed honestly as written, say so -- that
costs you nothing and is the correct answer.

## Task

- id: T-0001
- type: scaffold
- title: Repository bootstrap

## Acceptance criteria

Each criterion names the test that will be run to check it.

- AC-1 -- The editable install succeeds.
  checked by: smoke::pip_install

## Scope

Writable -- you may modify files matching:
- pyproject.toml

Creatable -- you may create new files matching:
- pyproject.toml

FROZEN -- you may not modify, create, or delete any file matching:
- none

## Rules

none -- no architecture decision records apply to this task's scope.

## Verification

redgear runs these commands itself after you exit and reads their exit codes.
Nothing you report about them is taken on trust.

- lint: ruff check --output-format=json --no-cache .

Gates run in this order and stop at the first failure:

1. scope_check
2. frozen_hash_check
3. lint

Not applied to a scaffold task: tests_pass, criteria_coverage, coverage_delta.

## Prior attempts

none -- this is attempt 1 of 3.

## Required outcome

When you are finished, return a JSON object matching the provided schema:

  outcome: "completed"          — you believe the task is done
           "blocked"            — you cannot proceed honestly
           "scope_insufficient" — the task requires editing a path you were not granted

  summary:        3 sentences maximum on what you changed and why
  changed_files:  every file you modified or created
  known_gaps:     anything knowingly incomplete

Reporting "blocked" or "scope_insufficient" does NOT count against your attempt
budget and is NOT a failure. It is the correct action when the task is ambiguous,
a dependency is missing, the environment is broken, or two rules contradict.

Declaring a gap in known_gaps is never penalised. Hiding one is caught by
verification and costs you an attempt.

Your work is verified independently after you exit: the git diff is recomputed,
frozen files are hash-checked, and the test suite is run by the orchestrator.
Claiming completion you cannot support will fail and consume an attempt.
