All pull requests

Pull requests

• Time five load-flaky tests by the product's own phases, not by start-up

  claude
  #58·solvyxtech opened Oct 8, 2026Owner·

  ·
  1

• Judge meter follow-ups; meter interview and mission planning as worker spend

  claude
  #57·solvyxtech opened Oct 8, 2026Owner·

  ·
  7

• Check the file before saying it has not changed since it was shown

  claude
  #55·solvyxtech opened Oct 8, 2026Owner·

  ·
  1

• Lean sessions: where long-session requests go, and flag-gated prototypes

  claude
  #50·solvyxtech opened Oct 8, 2026Owner·

  ·

• Lean guards: request-size ceilings in CI, per-run cost alarm in the bench

  claude
  #48·solvyxtech opened Oct 8, 2026Owner·

  ·
  5

• Refused tool calls stay cheap: excerpt on the wire, stop after six in a row

  claude
  #47·solvyxtech opened Oct 8, 2026Owner·

  ·
  2

• Exact tag: verified needs a check that compares the output with an exact value

  claude
  #46·solvyxtech opened Oct 8, 2026Owner·

  ·
  2

• Judge on Anthropic's API: read ANTHROPIC_API_KEY; ask without temperature when a model refuses it

  claude
  #45·solvyxtech opened Oct 8, 2026Owner·

  ·
  6

• Value tag: count behavioural assertions on literal input

  claude
  #44·solvyxtech opened Oct 8, 2026Owner·

  ·
  1

• Stop dropping good drafted checks; keep pre-work passes as refuse-only guards

  claude
  #43·solvyxtech opened Oct 8, 2026Owner·

  ·
  1

• Post-work audit: checks drafted after the work, gated on the pre-work tree and on mutants

  claude
  #42·solvyxtech opened Oct 8, 2026Owner·

  ·
  1

• Close the adversarial gaps with no PR yet

  claude
  #41·solvyxtech opened Oct 8, 2026Owner·

  ·
  1

• Keep check text and API keys away from the worker

  claude
  #40·solvyxtech opened Oct 8, 2026Owner·

  ·
  2

• test: adversarial regression suite for a hostile worker

  #39·solvyxtech opened Oct 8, 2026Owner·

  ·
  2

• Hidden-check commands off argv: stream them in on fd 3

  #38·solvyxtech opened Oct 8, 2026Owner·

  ·
  2

• review: cut receipt evidence at the real ## Output heading (worker could hide failing checks from the reviewer)

  #37·solvyxtech opened Oct 8, 2026Owner·

  ·
  2

• Send an output cap, and shrink it when the completion reserve overflows the window

  #36·solvyxtech opened Oct 7, 2026Owner·

  ·
  2

• Redraft checks that pass before the work; discrimination gate opt-in

  claude
  #34·solvyxtech opened Oct 7, 2026Owner·

  ·
  5

• Keep unattended workers on the task: bench isolation, scoped file tools, no-progress guard

  claude
  #33·solvyxtech opened Oct 7, 2026Owner·

  ·
  6

• Verified needs a check the worker did not write

  claude
  #32·solvyxtech opened Oct 7, 2026Owner·

  ·
  4

• Tool-call arguments go back to the provider as valid JSON

  claude
  #31·solvyxtech opened Oct 7, 2026Owner·

  ·
  2

• Context-overflow detection needs overflow wording

  claude
  #30·solvyxtech opened Oct 7, 2026Owner·

  ·
  2

• Run the worker's tools as another user: --worker-user (builds on #27)

  claude
  #29·solvyxtech opened Oct 7, 2026Owner·

  ·
  6

• Review objections must run: --review-executable

  claude
  #28·solvyxtech opened Oct 7, 2026Owner·

  ·
  3

• Keep hidden checks off disk during a job; task checks run in a throwaway copy

  claude
  #27·solvyxtech opened Oct 7, 2026Owner·

  ·
  6