coderace 0.8.0__tar.gz → 0.9.0__tar.gz

{coderace-0.8.0 → coderace-0.9.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: coderace
-Version: 0.8.0
+Version: 0.9.0
 Summary: Race coding agents against each other on real tasks
 Project-URL: Homepage, https://github.com/mikiships/coderace
 Project-URL: Repository, https://github.com/mikiships/coderace
@@ -106,6 +106,12 @@ description: |
 repo: .
 test_command: pytest tests/test_auth.py -x
 lint_command: ruff check .
+# Optional: independent verification suite written after agent completes
+# verify_command: python3 -m pytest verify_auth.py -x -q
+# verify_files:
+#   verify_auth.py: |
+#     def test_real_contract():
+#       assert True
 timeout: 300
 agents:
   - claude
@@ -133,6 +139,34 @@ For each agent in the task:
 | Wall time | 15% | Faster is better (normalized across agents) |
 | Lines changed | 10% | Fewer is better (normalized across agents) |
 
+## Verification Tests
+
+For stronger evaluation, tasks can define an independent verification suite that the agent does not control.
+
+```yaml
+verify_command: python3 -m pytest verify_api_contract.py -x -q
+verify_files:
+  verify_api_contract.py: |
+    def test_contract_behavior():
+      assert True
+```
+
+Flow for verification-enabled tasks:
+1. Agent completes implementation.
+2. `test_command` runs (agent-authored tests).
+3. `verify_files` are written into the workspace (overwriting same-path files).
+4. `verify_command` runs.
+
+Default scoring when `verify_command` is present:
+- tests: 25%
+- verify: 30%
+- exit: 20%
+- lint: 15%
+- time: 5%
+- lines: 5%
+
+Tasks without `verify_command` keep the legacy default scoring (40/20/15/15/10).
+
 ## Output
 
 Terminal table with Rich formatting:
@@ -172,6 +206,14 @@ coderace run --builtin fibonacci
 | `csv-analyzer` | Medium | CLI tool for CSV summary statistics |
 | `http-server` | Medium-Hard | HTTP/1.1 server using only stdlib socket |
 | `binary-search-tree` | Hard | AVL tree with insert, delete, search, and balancing |
+| `regex-engine` | Hard | Regex engine with custom matcher + verification suite |
+| `lru-cache` | Hard | Thread-safe LRU + TTL correctness verification |
+| `expression-evaluator` | Hard | Expression parser/evaluator with precedence and functions |
+| `url-router` | Hard | HTTP-style router with params, wildcard, and 405/404 logic |
+| `diff-algorithm` | Hard | Unified diff + patch application roundtrip checks |
+| `task-scheduler` | Hard | Dependency-aware priority scheduler with timeout handling |
+
+`coderace tasks list` now includes a `Verify` column so you can see which built-ins ship with verification suites.
 
 ## Try It Now
 

{coderace-0.8.0 → coderace-0.9.0}/README.md

@@ -76,6 +76,12 @@ description: |
 repo: .
 test_command: pytest tests/test_auth.py -x
 lint_command: ruff check .
+# Optional: independent verification suite written after agent completes
+# verify_command: python3 -m pytest verify_auth.py -x -q
+# verify_files:
+#   verify_auth.py: |
+#     def test_real_contract():
+#       assert True
 timeout: 300
 agents:
   - claude
@@ -103,6 +109,34 @@ For each agent in the task:
 | Wall time | 15% | Faster is better (normalized across agents) |
 | Lines changed | 10% | Fewer is better (normalized across agents) |
 
+## Verification Tests
+
+For stronger evaluation, tasks can define an independent verification suite that the agent does not control.
+
+```yaml
+verify_command: python3 -m pytest verify_api_contract.py -x -q
+verify_files:
+  verify_api_contract.py: |
+    def test_contract_behavior():
+      assert True
+```
+
+Flow for verification-enabled tasks:
+1. Agent completes implementation.
+2. `test_command` runs (agent-authored tests).
+3. `verify_files` are written into the workspace (overwriting same-path files).
+4. `verify_command` runs.
+
+Default scoring when `verify_command` is present:
+- tests: 25%
+- verify: 30%
+- exit: 20%
+- lint: 15%
+- time: 5%
+- lines: 5%
+
+Tasks without `verify_command` keep the legacy default scoring (40/20/15/15/10).
+
 ## Output
 
 Terminal table with Rich formatting:
@@ -142,6 +176,14 @@ coderace run --builtin fibonacci
 | `csv-analyzer` | Medium | CLI tool for CSV summary statistics |
 | `http-server` | Medium-Hard | HTTP/1.1 server using only stdlib socket |
 | `binary-search-tree` | Hard | AVL tree with insert, delete, search, and balancing |
+| `regex-engine` | Hard | Regex engine with custom matcher + verification suite |
+| `lru-cache` | Hard | Thread-safe LRU + TTL correctness verification |
+| `expression-evaluator` | Hard | Expression parser/evaluator with precedence and functions |
+| `url-router` | Hard | HTTP-style router with params, wildcard, and 405/404 logic |
+| `diff-algorithm` | Hard | Unified diff + patch application roundtrip checks |
+| `task-scheduler` | Hard | Dependency-aware priority scheduler with timeout handling |
+
+`coderace tasks list` now includes a `Verify` column so you can see which built-ins ship with verification suites.
 
 ## Try It Now
 

coderace-0.9.0/all-day-build-contract-v090-tasks.md

@@ -0,0 +1,163 @@
+# All-Day Build Contract: v0.9.0 — Verification Tests + New Tasks
+
+Status: In Progress
+Date: 2026-02-28
+Owner: Codex execution pass
+Scope type: Deliverable-gated (no hour promises)
+
+## 1. Objective
+
+Add verification tests to all 6 built-in tasks that lack them, and add 4 new built-in tasks that test different programming skills. The verification test runner (verify_command + verify_files) landed in v0.8.1 but only 6/12 tasks use it. Every task should have independent verification tests — tests the agent can't see, that validate the agent's implementation against a known-good spec. This is the feature that separates coderace from "run pytest and hope."
+
+New tasks should cover skills the current suite doesn't: file I/O, CLI argument parsing, data transformation pipeline, and a state machine. These are practical programming tasks that real developers encounter, not academic exercises.
+
+This contract is considered complete only when every deliverable and validation gate below is satisfied.
+
+## 2. Non-Negotiable Build Rules
+
+1. No time-based completion claims.
+2. Completion is allowed only when all checklist items are checked.
+3. Full test suite must pass at the end (`uv run python -m pytest`).
+4. New tasks must follow the exact YAML schema used by existing tasks.
+5. Never modify files outside the project directory.
+6. Commit after each completed deliverable (not at the end).
+7. If stuck on same issue for 3 attempts, stop and write a blocker report.
+8. Do NOT refactor, restyle, or "improve" code outside the deliverables.
+9. Read existing tasks and tests before writing new ones. Match style exactly.
+10. Every verify_files test must be self-contained (no imports from the task description that aren't part of the standard library + pytest).
+11. verify_command tests must be meaningfully different from test_command tests — they should test edge cases, integration behavior, or correctness properties the agent's own tests might miss.
+
+## 3. Feature Deliverables
+
+### D1. Add verify_command + verify_files to fibonacci.yaml
+
+The fibonacci task currently has no verification tests. Add verify_files with a verify test file that tests:
+- Large fibonacci numbers (fib(50), fib(100)) to ensure memoization actually works (not just that it doesn't crash)
+- Type checking: verify return types are int, not float
+- Edge case: fib(0) and fib(1) specifically
+- Performance: fib(100) should complete in under 1 second (verify memoization)
+- fibonacci_sequence returns a list, not a generator
+
+Required files to modify:
+- `coderace/builtins/tasks/fibonacci.yaml`
+
+- [ ] Add verify_command and verify_files to fibonacci.yaml
+- [ ] Verify tests are meaningfully different from the task's test_command expectations
+- [ ] Run `uv run python -m pytest tests/ -q` — all tests pass
+
+### D2. Add verify_command + verify_files to json-parser.yaml
+
+- [ ] Add verification tests that test malformed JSON edge cases, nested structures, unicode handling
+- [ ] Run `uv run python -m pytest tests/ -q` — all tests pass
+
+### D3. Add verify_command + verify_files to csv-analyzer.yaml
+
+- [ ] Add verification tests for empty CSVs, CSVs with quoted fields containing commas/newlines, large datasets
+- [ ] Run `uv run python -m pytest tests/ -q` — all tests pass
+
+### D4. Add verify_command + verify_files to markdown-to-html.yaml
+
+- [ ] Add verification tests for nested formatting, edge cases (empty input, only whitespace), HTML entity escaping
+- [ ] Run `uv run python -m pytest tests/ -q` — all tests pass
+
+### D5. Add verify_command + verify_files to http-server.yaml
+
+- [ ] Add verification tests for concurrent requests, proper HTTP headers, error responses, content-type handling
+- [ ] Run `uv run python -m pytest tests/ -q` — all tests pass
+
+### D6. Add verify_command + verify_files to binary-search-tree.yaml
+
+- [ ] Add verification tests for AVL balance property, deletion edge cases, large sequential inserts
+- [ ] Run `uv run python -m pytest tests/ -q` — all tests pass
+
+### D7. New task: file-watcher (medium)
+
+Create `coderace/builtins/tasks/file-watcher.yaml`:
+A CLI tool that watches a directory for file changes and logs them. Tests file I/O, directory traversal, and event handling.
+
+Requirements for the task description:
+- Implement `file_watcher.py` with a `FileWatcher` class
+- Methods: `scan(directory) -> dict[str, FileInfo]`, `diff(old_scan, new_scan) -> list[Change]`
+- FileInfo: path, size, modified_time, hash (md5)
+- Change types: added, modified, deleted
+- Must handle nested directories
+- Agent writes their own tests in `test_file_watcher.py`
+
+Include verify_command + verify_files from the start.
+
+- [ ] Create file-watcher.yaml with full task spec + verification tests
+- [ ] Register in builtins __init__.py if needed
+- [ ] Run `uv run python -m pytest tests/ -q` — all tests pass
+
+### D8. New task: cli-args-parser (medium)
+
+Create `coderace/builtins/tasks/cli-args-parser.yaml`:
+Build a lightweight CLI argument parser (like a mini argparse). Tests string parsing, API design, error handling.
+
+Requirements:
+- Implement `cli_parser.py` with an `ArgumentParser` class
+- Support: positional args, --flags, --key=value, --key value, -short flags, --no-flag (boolean negation)
+- Methods: `add_argument(name, type, default, required, help)`, `parse(args: list[str]) -> Namespace`
+- Proper error messages for missing required args, type conversion failures
+- Agent writes tests in `test_cli_parser.py`
+
+Include verify_command + verify_files.
+
+- [ ] Create cli-args-parser.yaml with full task spec + verification tests
+- [ ] Run `uv run python -m pytest tests/ -q` — all tests pass
+
+### D9. New task: data-pipeline (hard)
+
+Create `coderace/builtins/tasks/data-pipeline.yaml`:
+Build a composable data transformation pipeline. Tests functional programming patterns, method chaining, lazy evaluation.
+
+Requirements:
+- Implement `pipeline.py` with a `Pipeline` class
+- Methods: `map(fn)`, `filter(fn)`, `reduce(fn, initial)`, `sort(key)`, `take(n)`, `skip(n)`, `batch(size)`
+- Pipelines are lazy (don't execute until `.collect()` or `.first()`)
+- Pipelines are composable: `pipe1 | pipe2` creates a combined pipeline
+- Agent writes tests in `test_pipeline.py`
+
+Include verify_command + verify_files.
+
+- [ ] Create data-pipeline.yaml with full task spec + verification tests
+- [ ] Run `uv run python -m pytest tests/ -q` — all tests pass
+
+### D10. New task: state-machine (medium-hard)
+
+Create `coderace/builtins/tasks/state-machine.yaml`:
+Implement a finite state machine with transition guards and event handling. Tests OOP design, state management.
+
+Requirements:
+- Implement `state_machine.py` with `StateMachine`, `State`, `Transition` classes
+- Define states, transitions with guards (conditions), and actions (callbacks)
+- Methods: `add_state(name, on_enter, on_exit)`, `add_transition(from, to, event, guard)`, `trigger(event)`, `current_state`
+- Raise `InvalidTransition` for illegal transitions
+- Support `on_enter`/`on_exit` hooks for states
+- Agent writes tests in `test_state_machine.py`
+
+Include verify_command + verify_files.
+
+- [ ] Create state-machine.yaml with full task spec + verification tests
+- [ ] Run `uv run python -m pytest tests/ -q` — all tests pass
+
+## 4. Test Requirements
+
+- [ ] All existing 411 tests must still pass
+- [ ] New YAML tasks must be loadable by the task loader (test with `uv run python -c "from coderace.builtins import list_builtin_tasks; print(list_builtin_tasks())"`)
+- [ ] All new tasks must have valid YAML that passes task schema validation
+- [ ] verify_files content must be valid Python that would actually pass if a correct implementation existed
+- [ ] Final count: 16 built-in tasks, all with verify_command + verify_files
+
+## 5. Reports
+
+- Write progress to `progress-log.md` after each deliverable
+- Include: what was built, what tests pass, what's next, any blockers
+- Final summary when all deliverables done or stopped
+
+## 6. Stop Conditions
+
+- All deliverables checked and all tests passing -> DONE
+- 3 consecutive failed attempts on same issue -> STOP, write blocker report
+- Scope creep detected (new requirements discovered) -> STOP, report what's new
+- All tests passing but deliverables remain -> continue to next deliverable

coderace-0.9.0/all-day-build-contract-verification-tests.md

@@ -0,0 +1,165 @@
+# All-Day Build Contract: Verification Tests (v0.9.0)
+
+Status: In Progress
+Date: 2026-02-27
+Owner: Codex execution pass
+Scope type: Deliverable-gated (no hour promises)
+
+## 1. Objective
+
+Add "verification tests" to coderace: pre-written test suites that run AFTER an agent completes a task, validating the agent's implementation against a known standard. Currently agents score 100% because they write their own tests and naturally make them pass. Verification tests solve this by providing an independent quality check the agent doesn't control.
+
+New YAML fields:
+- `verify_command`: command to run verification tests (e.g., `python3 -m pytest verify_fibonacci.py -x -q`)
+- `verify_files`: dict of `{filename: content}` that get written into the task workspace BEFORE `verify_command` runs
+
+Scoring changes:
+- If task has `verify_command`: tests=25, verify=30, exit=20, lint=15, time=5, lines=5
+- If task has NO `verify_command`: scoring unchanged (backward compatible)
+
+This contract is considered complete only when every deliverable and validation gate below is satisfied.
+
+## 2. Non-Negotiable Build Rules
+
+1. No time-based completion claims.
+2. Completion is allowed only when all checklist items are checked.
+3. Full test suite must pass at the end.
+4. New features must ship with docs and report addendum updates in the same pass.
+5. CLI outputs must be deterministic and schema-backed where specified.
+6. Never modify files outside the project directory.
+7. Commit after each completed deliverable (not at the end).
+8. If stuck on same issue for 3 attempts, stop and write a blocker report.
+9. Do NOT refactor, restyle, or "improve" code outside the deliverables.
+10. Read existing tests and docs before writing new code.
+
+## 3. Feature Deliverables
+
+### D1. Verification Test Runner (core engine)
+
+Add verification test support to the coderace runner. After an agent completes a task and the normal `test_command` runs, if the task has `verify_command` and `verify_files`:
+1. Write each file from `verify_files` into the task workspace (overwriting any agent-written file with the same name is OK — that's the point)
+2. Run `verify_command` in the workspace
+3. Capture pass/fail/output the same way `test_command` is captured
+4. Add `verify_score`, `verify_output`, `verify_passed` to the result
+
+Required files to modify:
+- `coderace/runner.py` (or wherever task execution happens)
+- `coderace/models.py` or result dataclasses
+
+Read the existing runner code first to understand the execution flow.
+
+- [ ] Parse `verify_command` and `verify_files` from task YAML
+- [ ] Write verify_files to workspace after agent completes
+- [ ] Run verify_command and capture results
+- [ ] Add verify fields to result model
+- [ ] Tests for D1
+
+### D2. Scoring Engine Update
+
+Update the scoring engine to incorporate verification test results.
+
+When a task has `verify_command`:
+- `tests` weight: 25 (agent's own tests)
+- `verify` weight: 30 (verification tests)
+- `exit` weight: 20
+- `lint` weight: 15
+- `time` weight: 5
+- `lines` weight: 5
+
+When a task does NOT have `verify_command`:
+- Keep existing scoring unchanged (backward compatible)
+
+The task YAML can also override these weights with a `scoring:` section as it already does.
+
+Required files to modify:
+- `coderace/scoring.py` (or wherever scoring happens)
+
+- [ ] Add verify score calculation
+- [ ] Update weight distribution when verify_command present
+- [ ] Backward compatibility: no verify_command = old scoring
+- [ ] Tests for D2
+
+### D3. Benchmark Report Update
+
+Update the benchmark report (terminal, markdown, HTML) to show verification scores separately.
+
+In the results table, when any task has verification tests:
+- Add a "Verify" column showing verification pass/fail percentage
+- In task details, show verification output (truncated to 20 lines)
+
+Required files to modify:
+- `coderace/benchmark_report.py`
+- `coderace/commands/benchmark.py` (if CLI output changes)
+
+- [ ] Terminal report shows Verify column
+- [ ] Markdown report shows Verify column
+- [ ] HTML report shows Verify column
+- [ ] Verification output in task details
+- [ ] Tests for D3
+
+### D4. Six Hard Built-in Tasks with Verification Tests
+
+Create 6 new built-in tasks with embedded verification tests. These must be tasks where coding agents are likely to make mistakes. Each task YAML includes both a `test_command` (agents write their own tests) and `verify_command` + `verify_files` (our verification suite).
+
+Tasks to create in `coderace/builtins/tasks/`:
+
+1. **regex-engine** (hard): Implement a basic regex engine supporting `.`, `*`, `+`, `?`, `^`, `$`, character classes `[abc]`, `[a-z]`, `[^abc]`, and alternation `a|b`. Must handle backtracking correctly. Verification tests: catastrophic backtracking detection (must complete in <2s), nested quantifier edge cases, empty string matching.
+
+2. **lru-cache** (medium-hard): Implement an LRU cache with O(1) get/put, max capacity, TTL expiry, and thread safety. Verification tests: concurrent access correctness, TTL precision, capacity overflow ordering.
+
+3. **expression-evaluator** (hard): Implement a math expression parser and evaluator supporting +, -, *, /, **, parentheses, unary minus, variables, and functions (sin, cos, sqrt, abs). Must handle operator precedence correctly. Verification tests: deeply nested expressions, operator precedence edge cases, floating point precision, error handling for division by zero and undefined variables.
+
+4. **url-router** (medium-hard): Implement an HTTP URL router with path parameters (`/users/:id`), wildcards (`/files/*path`), middleware support, and method matching. Verification tests: ambiguous route resolution, parameter extraction, middleware ordering, 405 vs 404 distinction.
+
+5. **diff-algorithm** (hard): Implement Myers diff algorithm producing unified diff output. Must handle insertions, deletions, modifications, and generate minimal edit distance patches. Verification tests: large file diffs (1000+ lines), binary-like content, all-insert/all-delete edge cases, patch application roundtrip.
+
+6. **task-scheduler** (hard): Implement a priority-based task scheduler with dependencies (DAG), cycle detection, parallel execution slots, and timeout handling. Verification tests: diamond dependencies, cycle detection accuracy, priority inversion scenarios, timeout edge cases.
+
+Each task YAML must include:
+- Clear description (what to build)
+- `verify_files` with a comprehensive test file (30-50 test cases each)
+- `verify_command` pointing to the verification test file
+- Difficulty rating: hard
+- Scoring weights using the new verify-aware distribution
+
+- [ ] regex-engine.yaml with verify_files
+- [ ] lru-cache.yaml with verify_files
+- [ ] expression-evaluator.yaml with verify_files
+- [ ] url-router.yaml with verify_files
+- [ ] diff-algorithm.yaml with verify_files
+- [ ] task-scheduler.yaml with verify_files
+
+### D5. Documentation + Final Tests
+
+- Update README.md: Verification Tests section explaining the feature
+- Update `coderace tasks list` output to show which tasks have verification tests
+- Integration tests: run a mock task with verify_command, confirm scoring works end-to-end
+
+- [ ] README updated with Verification Tests section
+- [ ] `tasks list` shows verification badge
+- [ ] Integration test for full verify flow
+- [ ] All existing tests still pass
+
+## 4. Test Requirements
+
+- [ ] Unit tests for verify file writing
+- [ ] Unit tests for verify command execution
+- [ ] Unit tests for scoring with/without verify
+- [ ] Unit tests for report generation with verify columns
+- [ ] Integration test: task with verify_command scores differently than without
+- [ ] Integration test: verify_files correctly overwrite workspace files
+- [ ] Edge cases: verify_command fails, verify_files empty, task has verify but agent breaks the interface
+- [ ] All 392 existing tests must still pass
+
+## 5. Reports
+
+- Write progress to `progress-log.md` after each deliverable
+- Include: what was built, what tests pass, what's next, any blockers
+- Final summary when all deliverables done or stopped
+
+## 6. Stop Conditions
+
+- All deliverables checked and all tests passing -> DONE
+- 3 consecutive failed attempts on same issue -> STOP, write blocker report
+- Scope creep detected (new requirements discovered) -> STOP, report what's new
+- All tests passing but deliverables remain -> continue to next deliverable

coderace-0.9.0/benchmark-results/fibonacci-2026-02-27.md

@@ -0,0 +1,20 @@
+# coderace Benchmark Results
+
+Generated: 2026-02-27 15:37 UTC
+
+Benchmark ID: `bench-20260227-153601`
+
+| Task | claude | codex |
+|------|------|------|
+| fibonacci | 35.0 (34s) | 35.0 (62s) |
+|------|------|------|
+| **TOTAL** | **35.0** | **35.0** |
+| **Win Rate** | 100% | 100% |
+| **Avg Time** | 33.6s | 61.6s |
+| **Total Cost** | $0.0218 | - |
+
+## Task Insights
+
+| Task | Best Agent | Best Score | Avg Score | Fastest Agent |
+|------|-----------|-----------|----------|--------------|
+| fibonacci | claude | 35.0 | 35.0 | claude (34s) |

coderace-0.9.0/benchmark-results/fibonacci-v2-2026-02-27.md

@@ -0,0 +1,20 @@
+# coderace Benchmark Results
+
+Generated: 2026-02-27 15:40 UTC
+
+Benchmark ID: `bench-20260227-153946`
+
+| Task | claude | codex |
+|------|------|------|
+| fibonacci | 100.0 (34s) | 100.0 (26s) |
+|------|------|------|
+| **TOTAL** | **100.0** | **100.0** |
+| **Win Rate** | 100% | 100% |
+| **Avg Time** | 33.7s | 25.9s |
+| **Total Cost** | $0.0173 | - |
+
+## Task Insights
+
+| Task | Best Agent | Best Score | Avg Score | Fastest Agent |
+|------|-----------|-----------|----------|--------------|
+| fibonacci | claude | 100.0 | 100.0 | codex (26s) |

coderace-0.9.0/benchmark-results/hard-tasks-2026-02-27.md

@@ -0,0 +1,24 @@
+# coderace Benchmark Results
+
+Generated: 2026-02-27 15:55 UTC
+
+Benchmark ID: `bench-20260227-154647`
+
+| Task | claude | codex |
+|------|------|------|
+| binary-search-tree | 100.0 (47s) | 100.0 (33s) |
+| csv-analyzer | 100.0 (71s) | 100.0 (137s) |
+| http-server | 100.0 (64s) | 100.0 (185s) |
+|------|------|------|
+| **TOTAL** | **300.0** | **300.0** |
+| **Win Rate** | 100% | 100% |
+| **Avg Time** | 60.8s | 118.2s |
+| **Total Cost** | $0.1669 | - |
+
+## Task Insights
+
+| Task | Best Agent | Best Score | Avg Score | Fastest Agent |
+|------|-----------|-----------|----------|--------------|
+| binary-search-tree | claude | 100.0 | 100.0 | codex (33s) |
+| csv-analyzer | claude | 100.0 | 100.0 | claude (71s) |
+| http-server | claude | 100.0 | 100.0 | claude (64s) |

coderace-0.9.0/benchmark-results/multi-task-2026-02-27.md

@@ -0,0 +1,24 @@
+# coderace Benchmark Results
+
+Generated: 2026-02-27 15:46 UTC
+
+Benchmark ID: `bench-20260227-154058`
+
+| Task | claude | codex |
+|------|------|------|
+| fibonacci | 100.0 (25s) | 100.0 (71s) |
+| json-parser | 100.0 (66s) | 100.0 (42s) |
+| markdown-to-html | 100.0 (74s) | 100.0 (48s) |
+|------|------|------|
+| **TOTAL** | **300.0** | **300.0** |
+| **Win Rate** | 100% | 100% |
+| **Avg Time** | 55.1s | 53.5s |
+| **Total Cost** | $0.1471 | - |
+
+## Task Insights
+
+| Task | Best Agent | Best Score | Avg Score | Fastest Agent |
+|------|-----------|-----------|----------|--------------|
+| fibonacci | claude | 100.0 | 100.0 | claude (25s) |
+| json-parser | claude | 100.0 | 100.0 | codex (42s) |
+| markdown-to-html | claude | 100.0 | 100.0 | codex (48s) |

{coderace-0.8.0 → coderace-0.9.0}/coderace/adapters/codex.py

@@ -16,9 +16,8 @@ class CodexAdapter(BaseAdapter):
     def build_command(self, task_description: str) -> list[str]:
         return [
             "codex",
-            "--quiet",
+            "exec",
             "--full-auto",
-            "-p",
             task_description,
         ]