Commits

Branch selector

master

User selector

All users

All time

Commit history

Commits on Jul 2, 2026

• chore: add local agent skills and update skills-lock

  av

  committed
  766d70bView commit details

  Copy full SHA for 766d70b

  Browse repository at this point

• feat: 22-task discriminative TB 2.1 subset runner + full-run analysis docs

  Show description for d51483e
  av

  committed
  d51483eView commit details

  Copy full SHA for d51483e

  Browse repository at this point

Commits on Jul 1, 2026

• fix: worker restarts stateful services cleanly before final ACK

  av

  committed
  bd9b482View commit details

  Copy full SHA for bd9b482

  Browse repository at this point

• fix: skeptical recheck may NACK only on a concrete failing criterion

  av

  committed
  6ed0f5bView commit details

  Copy full SHA for 6ed0f5b

  Browse repository at this point

• fix: non-destructive salvage — preserve passing artifacts, minimal repairs only

  av

  committed
  7f2e6ecView commit details

  Copy full SHA for 7f2e6ec

  Browse repository at this point

• fix: judge verifies exact thresholds — remove 1.05x margin that caused false NACKs

  av

  committed
  78934d1View commit details

  Copy full SHA for 78934d1

  Browse repository at this point

Commits on Jun 27, 2026

• fix: enforce delegate timeout with SIGKILL and done-flag guard

  Show description for 42f185c
  av

  committed
  42f185cView commit details

  Copy full SHA for 42f185c

  Browse repository at this point

• fix: compute per-task timeout from Harbor's cached task.toml

  Show description for 13b9744
  av

  committed
  13b9744View commit details

  Copy full SHA for 13b9744

  Browse repository at this point

• fix: remove hardcoded MI_TASK_TIMEOUT=900s default — miscalibrates budget phases

  Show description for 0b91595
  av

  committed
  0b91595View commit details

  Copy full SHA for 0b91595

  Browse repository at this point

• fix: eval-optimized system prompt + README injection + stronger TERMINAL_BENCH_CHECK

  Show description for c66adb2
  av

  committed
  c66adb2View commit details

  Copy full SHA for c66adb2

  Browse repository at this point

• fix: report actual iteration count on goal NACK exit instead of limit constant

  av

  committed
  97fa1b4View commit details

  Copy full SHA for 97fa1b4

  Browse repository at this point

• fix: judge runs /tests/ first as ground truth — prevents false-ACK on 4 flaky task patterns

  Show description for 4eb1772
  av

  committed
  4eb1772View commit details

  Copy full SHA for 4eb1772

  Browse repository at this point

• fix: nohup for bg services + judge anti-kill directive

  Show description for a1e3113
  av

  committed
  a1e3113View commit details

  Copy full SHA for a1e3113

  Browse repository at this point

• fix: improve workdir detection — add /workdir, skip empty dirs, BusyBox compat

  Show description for 0e8b263
  av

  committed
  0e8b263View commit details

  Copy full SHA for 0e8b263

  Browse repository at this point

• fix: enable budget-aware goal loop by defaulting MI_TASK_TIMEOUT to 900s

  Show description for d0dc249
  av

  committed
  d0dc249View commit details

  Copy full SHA for d0dc249

  Browse repository at this point

• fix: retry transient mid-stream API errors + recover from malformed tool args

  Show description for d481673
  av

  committed
  d481673View commit details

  Copy full SHA for d481673

  Browse repository at this point

• fix: cap planner at 15% of budget + streamline discovery from 8 to 5 steps

  Show description for f261f16
  av

  committed
  f261f16View commit details

  Copy full SHA for f261f16

  Browse repository at this point

• refactor: compress worker prompt 65% — remove coding-specific directives redundant with planner

  Show description for db0a15e
  av

  committed
  db0a15eView commit details

  Copy full SHA for db0a15e

  Browse repository at this point

• fix: strengthen test-first discovery + coverage enumeration + judge runs verifier tests

  Show description for 6c2aac0
  av

  committed
  6c2aac0View commit details

  Copy full SHA for 6c2aac0

  Browse repository at this point

• fix: judge anti-side-effect guard + worker anti-self-sabotage directives

  Show description for c038155
  av

  committed
  c038155View commit details

  Copy full SHA for c038155

  Browse repository at this point

• fix: smart plan truncation (head+tail) + extend skeptical recheck to iterations 1-2

  Show description for d8bd066
  av

  committed
  d8bd066View commit details

  Copy full SHA for d8bd066

  Browse repository at this point

• fix: goal loop prompt quality — compress worker directives, fix planner test patterns

  Show description for 1d7968b
  av

  committed
  1d7968bView commit details

  Copy full SHA for 1d7968b

  Browse repository at this point

• fix: cap first goal iteration to 1/3 of time budget to prevent premature convergence

  Show description for 20316e9
  av

  committed
  20316e9View commit details

  Copy full SHA for 20316e9

  Browse repository at this point

• fix: planner test discovery, anti-fabrication guard, Docker process management

  Show description for 4f3cadb
  av

  committed
  4f3cadbView commit details

  Copy full SHA for 4f3cadb

  Browse repository at this point

• fix: test-first worker strategy + skeptical judge on first-try ACKs

  Show description for 6172b66
  av

  committed
  6172b66View commit details

  Copy full SHA for 6172b66

  Browse repository at this point

Commits on Jun 26, 2026

• test: add coverage for blocker detection, spawn error skip, forced salvage, bash truncation

  av

  committed
  fa8b9deView commit details

  Copy full SHA for fa8b9de

  Browse repository at this point

• fix: bash tool truncates output at 50KB to prevent context overflow

  av

  committed
  2ad069cView commit details

  Copy full SHA for 2ad069c

  Browse repository at this point

• fix: skip judge call on worker spawn/exit/timeout errors to save budget

  av

  committed
  20b9fcdView commit details

  Copy full SHA for 20b9fcd

  Browse repository at this point

• refactor: consolidate goal loop prompts — 46% token reduction

  Show description for 10082c5
  av

  committed
  10082c5View commit details

  Copy full SHA for 10082c5

  Browse repository at this point

• fix: forced salvage on 5+ consecutive identical blockers in deadline mode + 402 test

  Show description for 0de40f8
  av

  committed
  0de40f8View commit details

  Copy full SHA for 0de40f8

  Browse repository at this point

• feat: worker requires strategy differentiation and two-phase last-mile repair

  av

  committed
  cd11140View commit details

  Copy full SHA for cd11140

  Browse repository at this point

• feat: repeated-blocker detection forces approach change after 3 identical NACKs

  av

  committed
  0f4af50View commit details

  Copy full SHA for 0f4af50

  Browse repository at this point

• feat: SOTA-informed improvements — test source reading, service detection, COMMIT phase lock

  Show description for 2319bba
  av

  committed
  2319bbaView commit details

  Copy full SHA for 2319bba

  Browse repository at this point

• feat: judge reports failing test names on near-complete NACK

  Show description for 61dd32f
  av

  committed
  61dd32fView commit details

  Copy full SHA for 61dd32f

  Browse repository at this point

• feat: worker prompt adds last-mile test focus and service management

  Show description for 5df159d
  av

  committed
  5df159dView commit details

  Copy full SHA for 5df159d

  Browse repository at this point