graff — a fast agentic coding harness in Zig: multi-provider, MCP, workflows, DGM evolution loop, TS/Python SDKs
Python
{"task": "openssl-selfsigned-cert", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/openssl-selfsigned-cert:20251031", "wall_s": 70.3, "agent_exit": 0, "agent_tail": ".company.local\n- `/app/ssl/server.pem` \u2014 key + cert combined\n- `/app/ssl/verification.txt` \u2014 subject, validity dates, SHA-256 fingerprint\n- `/app/check_cert.py` \u2014 loads the cert and prints CN, expiration (YYYY-MM-DD), then `Certificate verification successful`\n\nKey and cert modulus match; checker output: CN `dev-internal.company.local`, expires `2027-09-03`.\n[usage] 7 api call(s) \u00b7 45377 in (27136 cached, 0 cache writes) + 4779 out tokens \u00b7 $0.0000 \u00b7 7 subscription call(s), flat-rate (not in $)\n", "calls": 7, "tok_in": 45377, "tok_cached": 27136, "tok_out": 4779, "cost_usd": 0.0, "verify_tail": "...... [100%]\n6 passed in 0.93s\nVEXIT:0", "pass_": true, "list_ordinary": 18241, "list_cached": 27136, "list_out": 4779, "list_prompt": 45377, "list_tokens": 50156, "list_token_usd": 0.078724, "list_tool_usd": 0.0, "list_usd": 0.078724, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.598, "tok_per_call": 6482}
{"task": "build-cython-ext", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/build-cython-ext:20251031", "wall_s": 723.9, "agent_exit": 0, "agent_tail": "d from site-packages:\n\n- `pyknotid.spacecurves.chelpers`\n- `pyknotid.spacecurves.ccomplexity`\n- `pyknotid.cinvariants`\n\n[boot +293ms, phase 293ms] credentials/model\ncalling grok-4.6\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n [main] \u21af 2 failed\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n [main] \u21af 1 failed\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n [main] \u21af 1 failed\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n[usage] 53 api call(s) \u00b7 1950257 in (1682688 cached, 0 cache writes) + 12587 out tokens \u00b7 $0.0000 \u00b7 53 subscription call(s), flat-rate (not in $)\n", "calls": 53, "tok_in": 1950257, "tok_cached": 1682688, "tok_out": 12587, "cost_usd": 0.0, "verify_tail": "........... [100%]\n11 passed in 11.92s\nVEXIT:0", "pass_": true, "list_ordinary": 267569, "list_cached": 1682688, "list_out": 12587, "list_prompt": 1950257, "list_tokens": 1962844, "list_token_usd": 2.904008, "list_tool_usd": 0.0, "list_usd": 1.452004, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.8628, "tok_per_call": 36797}
{"task": "chess-best-move", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/chess-best-move:20251031", "wall_s": 405.0, "agent_exit": 0, "agent_tail": "the exposed king on f5. Both mates are in `/app/move.txt`:\n\n- e2e4 (Qe4#)\n- g2g4 (g4#)\n\n[boot +308ms, phase 308ms] credentials/model\ncalling grok-4.6\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n [main] \u21af 2 failed\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n [main] White is mating in one from the exposed king on f5. Both mates are in `/app/move.txt`:\n\n- e2e4 (Qe4#)\n- g2g4 (g4#)\n[usage] 30 api call(s) \u00b7 602510 in (509312 cached, 0 cache writes) + 23136 out tokens \u00b7 $0.0000 \u00b7 30 subscription call(s), flat-rate (not in $)\n", "calls": 30, "tok_in": 602510, "tok_cached": 509312, "tok_out": 23136, "cost_usd": 0.0, "verify_tail": ". [100%]\n1 passed in 0.03s\nVEXIT:0", "pass_": true, "list_ordinary": 93198, "list_cached": 509312, "list_out": 23136, "list_prompt": 602510, "list_tokens": 625646, "list_token_usd": 1.159736, "list_tool_usd": 0.0, "list_usd": 0.579868, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.8453, "tok_per_call": 20084}
{"task": "code-from-image", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/code-from-image:20251031", "wall_s": 151.4, "agent_exit": 0, "agent_tail": "3b9ecda444c70ec22cafef6e31a3de7af6d047974dc90ce3defe`\n\n[boot +301ms, phase 301ms] credentials/model\ncalling grok-4.6\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n [main] Implemented the snippet: SHA-256 of `code.png`, then SHA-256(`h0 || h0[:10] || \"0000TBENCH-SALT\"`). Wrote the hex digest to `/app/output.txt`:\n\n`bee26a133f103b9ecda444c70ec22cafef6e31a3de7af6d047974dc90ce3defe`\n[usage] 11 api call(s) \u00b7 54435 in (45952 cached, 0 cache writes) + 2106 out tokens \u00b7 $0.0000 \u00b7 11 subscription call(s), flat-rate (not in $)\n", "calls": 11, "tok_in": 54435, "tok_cached": 45952, "tok_out": 2106, "cost_usd": 0.0, "verify_tail": ".. [100%]\n2 passed in 0.03s\nVEXIT:0", "pass_": true, "list_ordinary": 8483, "list_cached": 45952, "list_out": 2106, "list_prompt": 54435, "list_tokens": 56541, "list_token_usd": 0.052578, "list_tool_usd": 0.0, "list_usd": 0.052578, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.8442, "tok_per_call": 4949}
{"task": "constraints-scheduling", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/constraints-scheduling:20251031", "wall_s": 103.8, "agent_exit": 0, "agent_tail": "ob not before 10:00, Carol 09:00\u201317:00 except lunch 12:00\u201312:30, plus existing events.\n\nMon/Tue have no common free hour. Next valid time is Thu 11:00; Wed 11:00 is earlier and is in Alice\u2019s morning window (Carol Monday preference does not apply).\n\nInput calendars were not modified.\n\n[boot +304ms, phase 304ms] credentials/model\ncalling grok-4.6\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n[usage] 7 api call(s) \u00b7 68185 in (55424 cached, 0 cache writes) + 6179 out tokens \u00b7 $0.0000 \u00b7 7 subscription call(s), flat-rate (not in $)\n", "calls": 7, "tok_in": 68185, "tok_cached": 55424, "tok_out": 6179, "cost_usd": 0.0, "verify_tail": "... [100%]\n3 passed in 3.38s\nVEXIT:0", "pass_": true, "list_ordinary": 12761, "list_cached": 55424, "list_out": 6179, "list_prompt": 68185, "list_tokens": 74364, "list_token_usd": 0.090308, "list_tool_usd": 0.0, "list_usd": 0.090308, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.8128, "tok_per_call": 9741}
{"task": "log-summary-date-ranges", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/log-summary-date-ranges:20251031", "wall_s": 24.9, "agent_exit": 0, "agent_tail": " ignored.\n\nperiod,severity,count\ntoday,ERROR,370\ntoday,WARNING,463\ntoday,INFO,1433\nlast_7_days,ERROR,2969\nlast_7_days,WARNING,4020\nlast_7_days,INFO,12203\nlast_30_days,ERROR,9594\nlast_30_days,WARNING,12847\nlast_30_days,INFO,38702\nmonth_to_date,ERROR,4682\nmonth_to_date,WARNING,6327\nmonth_to_date,INFO,19420\ntotal,ERROR,14160\ntotal,WARNING,18772\ntotal,INFO,56621\n[usage] 3 api call(s) \u00b7 26865 in (17664 cached, 0 cache writes) + 1676 out tokens \u00b7 $0.0000 \u00b7 3 subscription call(s), flat-rate (not in $)\n", "calls": 3, "tok_in": 26865, "tok_cached": 17664, "tok_out": 1676, "cost_usd": 0.0, "verify_tail": ".. [100%]\n2 passed in 0.04s\nVEXIT:0", "pass_": true, "list_ordinary": 9201, "list_cached": 17664, "list_out": 1676, "list_prompt": 26865, "list_tokens": 28541, "list_token_usd": 0.03729, "list_tool_usd": 0.0, "list_usd": 0.03729, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.6575, "tok_per_call": 8955}
{"task": "modernize-scientific-stack", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/modernize-scientific-stack:20251031", "wall_s": 26.7, "agent_exit": 0, "agent_tail": "] Modernized the climate analyzer for Python 3.\n\n`/app/analyze_climate_modern.py` reads the CSV with pandas (UTF-8), uses pathlib + configparser, and prints:\n- Station 101 mean temperature: -15.5\u00b0C\n- Station 102 mean temperature: 30.3\u00b0C\n\n`/app/requirements.txt` pins numpy, pandas, matplotlib, and scipy with `>=` constraints. Legacy files were left unchanged.\n[usage] 6 api call(s) \u00b7 35357 in (10112 cached, 0 cache writes) + 1416 out tokens \u00b7 $0.0000 \u00b7 6 subscription call(s), flat-rate (not in $)\n", "calls": 6, "tok_in": 35357, "tok_cached": 10112, "tok_out": 1416, "cost_usd": 0.0, "verify_tail": ".. [100%]\n2 passed in 1.53s\nVEXIT:0", "pass_": true, "list_ordinary": 25245, "list_cached": 10112, "list_out": 1416, "list_prompt": 35357, "list_tokens": 36773, "list_token_usd": 0.064042, "list_tool_usd": 0.0, "list_usd": 0.064042, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.286, "tok_per_call": 5893}
{"task": "multi-source-data-merger", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/multi-source-data-merger:20251031", "wall_s": 63.6, "agent_exit": 0, "agent_tail": "s)\n- **3 conflicts** on user 101: `name`, `email`, `created_date`\n\n[boot +305ms, phase 305ms] credentials/model\ncalling grok-4.6\n\u203a\n\u203a\n\u203a\n [main] Merged users from JSON/CSV/Parquet into `/app/merged_users.parquet` (4 rows) and `/app/conflicts.json` (3 conflicts on user 101). Field aliases mapped; source_a wins on conflicts; dates are YYYY-MM-DD; user_id is int.\n[usage] 3 api call(s) \u00b7 18245 in (6656 cached, 0 cache writes) + 4460 out tokens \u00b7 $0.0000 \u00b7 3 subscription call(s), flat-rate (not in $)\n", "calls": 3, "tok_in": 18245, "tok_cached": 6656, "tok_out": 4460, "cost_usd": 0.0, "verify_tail": "... [100%]\n3 passed in 2.00s\nVEXIT:0", "pass_": true, "list_ordinary": 11589, "list_cached": 6656, "list_out": 4460, "list_prompt": 18245, "list_tokens": 22705, "list_token_usd": 0.053266, "list_tool_usd": 0.0, "list_usd": 0.053266, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.3648, "tok_per_call": 6082}
{"task": "sqlite-db-truncate", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/sqlite-db-truncate:20251031", "wall_s": 34.6, "agent_exit": 0, "agent_tail": "e 4KB page.\n\n[boot +306ms, phase 306ms] credentials/model\ncalling grok-4.6\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n [main] Recovered 10 rows from the truncated SQLite leaf page in `/app/trunc.db` (page type 0x0d, 10 cells) and wrote them to `/app/recover.json`.\n\nRows: testword00\u201309 with values 1, 2, 10, 25, 42, 50, 75, 99, 99.99, 0.5. No other cell payloads were present on the 4KB page.\n[usage] 5 api call(s) \u00b7 28900 in (16256 cached, 0 cache writes) + 2395 out tokens \u00b7 $0.0000 \u00b7 5 subscription call(s), flat-rate (not in $)\n", "calls": 5, "tok_in": 28900, "tok_cached": 16256, "tok_out": 2395, "cost_usd": 0.0, "verify_tail": ". [100%]\n1 passed in 0.03s\nVEXIT:0", "pass_": true, "list_ordinary": 12644, "list_cached": 16256, "list_out": 2395, "list_prompt": 28900, "list_tokens": 31295, "list_token_usd": 0.047786, "list_tool_usd": 0.0, "list_usd": 0.047786, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.5625, "tok_per_call": 5780}
{"task": "git-leak-recovery", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/git-leak-recovery:20251031", "wall_s": 27.8, "agent_exit": 0, "agent_tail": "ms] credentials/model\ncalling grok-4.6\n\u203a\n [main] \u21af 1 failed\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n [main] Recovered `secret[lost_and_found_in_git]` from dangling commit `6b4c75c` (reset via reflog) and wrote it to `/app/secret.txt`.\n\nScrubbed the repo with `git reflog expire` + `git gc --prune=now`. The secret blob/commit/tree are gone; remaining commits and messages are unchanged.\n[usage] 6 api call(s) \u00b7 31404 in (27520 cached, 0 cache writes) + 1490 out tokens \u00b7 $0.0000 \u00b7 6 subscription call(s), flat-rate (not in $)\n", "calls": 6, "tok_in": 31404, "tok_cached": 27520, "tok_out": 1490, "cost_usd": 0.0, "verify_tail": "py:101\n /tests/test_outputs.py:101: SyntaxWarning: invalid escape sequence '\\.'\n \"find /app/repo -type f -not -path '*/\\.*' \"\n\n-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html\n5 passed, 1 warning in 0.30s\nVEXIT:0", "pass_": true, "list_ordinary": 3884, "list_cached": 27520, "list_out": 1490, "list_prompt": 31404, "list_tokens": 32894, "list_token_usd": 0.030468, "list_tool_usd": 0.0, "list_usd": 0.030468, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.8763, "tok_per_call": 5234}
{"task": "regex-log", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/regex-log:20251031", "wall_s": 140.4, "agent_exit": 0, "agent_tail": "del\ncalling grok-4.6\n\u203a\n [main] \u21af 1 failed\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n [main] Regex saved in /app/regex.txt:1.\n\nIt requires an IPv4 on the same line (octets 0\u2013255, no leading zeros, not touching alphanumerics), then captures only the last valid YYYY-MM-DD on that line (Feb 01\u201329; 30/31 by month; same bounds). One capturing group so re.findall returns the date strings.\n[usage] 8 api call(s) \u00b7 82702 in (58368 cached, 0 cache writes) + 9743 out tokens \u00b7 $0.0000 \u00b7 8 subscription call(s), flat-rate (not in $)\n", "calls": 8, "tok_in": 82702, "tok_cached": 58368, "tok_out": 9743, "cost_usd": 0.0, "verify_tail": ". [100%]\n1 passed in 0.03s\nVEXIT:0", "pass_": true, "list_ordinary": 24334, "list_cached": 58368, "list_out": 9743, "list_prompt": 82702, "list_tokens": 92445, "list_token_usd": 0.13631, "list_tool_usd": 0.0, "list_usd": 0.13631, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.7058, "tok_per_call": 10338}
{"task": "sanitize-git-repo", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/sanitize-git-repo:20251031", "wall_s": 144.1, "agent_exit": 0, "agent_tail": "ing Face token\n- `ray_processing/process.py` \u2014 AWS keys\n- `exp_data/datasets/tokenized/rw_v2_fasttext_openhermes_vs_rw_v2_bigram_0.1_arcade100k.json` \u2014 Hugging Face token (in a stored git diff)\n\nA follow-up scan found none of the original secret values remaining in tracked files. Placeholders are consistent across the repo. Git history was not rewritten.\n[usage] 17 api call(s) \u00b7 331603 in (216576 cached, 0 cache writes) + 6242 out tokens \u00b7 $0.0000 \u00b7 17 subscription call(s), flat-rate (not in $)\n", "calls": 17, "tok_in": 331603, "tok_cached": 216576, "tok_out": 6242, "cost_usd": 0.0, "verify_tail": "... [100%]\n3 passed in 1.02s\nVEXIT:0", "pass_": true, "list_ordinary": 115027, "list_cached": 216576, "list_out": 6242, "list_prompt": 331603, "list_tokens": 337845, "list_token_usd": 0.751588, "list_tool_usd": 0.0, "list_usd": 0.375794, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.6531, "tok_per_call": 19506}
{"task": "dna-insert", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/dna-insert:20251031", "wall_s": 182.9, "agent_exit": 0, "agent_tail": " (input pos 213).\n\n`primers.fasta`:\n- forward (5\u2032 insertion + 16 nt anneal, Tm 63.48\u00b0C): AGTAGATTAGAAGAAGAATTAAGAAGAAGATTAACAGAAAGCAAGGGCGAGGAGC\n- reverse (44 nt anneal, Tm 63.17\u00b0C): CATATGTATATCTCCTTCTTAAAGTTAAACAAAATTATTTCTAG\n\nAnnealing lengths 16/44 (15\u201345), Tms within 58\u201372\u00b0C and 0.31\u00b0C of each other (oligotm `-tp 1 -sc 1 -mv 50 -dv 2 -n 0.8 -d 500`).\n[usage] 15 api call(s) \u00b7 147737 in (89600 cached, 0 cache writes) + 7677 out tokens \u00b7 $0.0000 \u00b7 15 subscription call(s), flat-rate (not in $)\n", "calls": 15, "tok_in": 147737, "tok_cached": 89600, "tok_out": 7677, "cost_usd": 0.0, "verify_tail": ". [100%]\n1 passed in 0.10s\nVEXIT:0", "pass_": true, "list_ordinary": 58137, "list_cached": 89600, "list_out": 7677, "list_prompt": 147737, "list_tokens": 155414, "list_token_usd": 0.207136, "list_tool_usd": 0.0, "list_usd": 0.207136, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.6065, "tok_per_call": 9849}
{"task": "merge-diff-arc-agi-task", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/merge-diff-arc-agi-task:20251031", "wall_s": 141.9, "agent_exit": 0, "agent_tail": "-4.6\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n [main] Repo is at `/app/repo` with `branch1` and `branch2` fetched from the two bundles. `branch2` is merged into `branch1`.\n\n`/app/repo/algo.py` implements `map`: non-zero cells define a period-3 anti-diagonal tiling `(i+j) % 3`, then the grid is filled with that pattern. All three cases in `/app/examples.json` match exactly.\n[usage] 11 api call(s) \u00b7 132388 in (88704 cached, 0 cache writes) + 4212 out tokens \u00b7 $0.0000 \u00b7 11 subscription call(s), flat-rate (not in $)\n", "calls": 11, "tok_in": 132388, "tok_cached": 88704, "tok_out": 4212, "cost_usd": 0.0, "verify_tail": "..... [100%]\n5 passed in 0.10s\nVEXIT:0", "pass_": true, "list_ordinary": 43684, "list_cached": 88704, "list_out": 4212, "list_prompt": 132388, "list_tokens": 136600, "list_token_usd": 0.156992, "list_tool_usd": 0.0, "list_usd": 0.156992, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.67, "tok_per_call": 12035}
{"task": "db-wal-recovery", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/db-wal-recovery:20251031", "wall_s": 357.8, "agent_exit": 0, "agent_tail": "510ms, phase 510ms] credentials/model\ncalling grok-4.6\n\u203a\n [main] \u21af 1 failed\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n [main] Recovered all 11 rows into `/app/recovered.json`. The WAL was XOR\u2019d with `0x42`; after decrypting it, SQLite applied the logged updates (`apple`/`banana` \u2192 150/250) and inserts (`fig`\u2013`mango`, 600\u20131100).\n[usage] 32 api call(s) \u00b7 1415071 in (1195776 cached, 0 cache writes) + 19856 out tokens \u00b7 $0.0000 \u00b7 32 subscription call(s), flat-rate (not in $)\n", "calls": 32, "tok_in": 1415071, "tok_cached": 1195776, "tok_out": 19856, "cost_usd": 0.0, "list_ordinary": 219295, "list_cached": 1195776, "list_out": 19856, "list_prompt": 1415071, "list_tokens": 1434927, "list_token_usd": 2.311228, "list_tool_usd": 0.0, "list_usd": 1.155614, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.845, "tok_per_call": 44221, "verify_tail": "....... [100%]\n7 passed in 0.11s\nVEXIT:0", "pass_": true}
{"task": "polyglot-c-py", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/polyglot-c-py:20251031", "wall_s": 122.8, "agent_exit": 0, "agent_tail": "als/model\ncalling grok-4.6\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n [main] Polyglot is at polyglot/main.py.c. C sees a block comment around the Python (`#define _ /*` \u2026 `*/`), Python sees a `#` comment then a `\"\"\"` string around the C.\n\nBoth take N from argv and print f(N) with f(0)=0, f(1)=1 (iterative). Verified for 0, 1, 2, 10, 20. Directory contains only that file.\n[usage] 13 api call(s) \u00b7 94120 in (86400 cached, 0 cache writes) + 4768 out tokens \u00b7 $0.0000 \u00b7 13 subscription call(s), flat-rate (not in $)\n", "calls": 13, "tok_in": 94120, "tok_cached": 86400, "tok_out": 4768, "cost_usd": 0.0, "list_ordinary": 7720, "list_cached": 86400, "list_out": 4768, "list_prompt": 94120, "list_tokens": 98888, "list_token_usd": 0.087248, "list_tool_usd": 0.0, "list_usd": 0.087248, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.918, "tok_per_call": 7240, "verify_tail": ". [100%]\n1 passed in 0.97s\nVEXIT:0", "pass_": true}
{"task": "vulnerable-secret", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/vulnerable-secret:20251031", "wall_s": 32.5, "agent_exit": 0, "agent_tail": "l0w_m4st3r_k3y_2024} from XOR-encoded data in /app/vulnerable (.data @ 0x404060, key 0x42) and wrote it to /app/results.txt.\n\n[boot +297ms, phase 297ms] credentials/model\ncalling grok-4.6\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n [main] Extracted FLAG{b4ff3r_0v3rfl0w_m4st3r_k3y_2024} from XOR-encoded data in /app/vulnerable (.data @ 0x404060, key 0x42) and wrote it to /app/results.txt.\n[usage] 6 api call(s) \u00b7 54302 in (38016 cached, 0 cache writes) + 2658 out tokens \u00b7 $0.0000 \u00b7 6 subscription call(s), flat-rate (not in $)\n", "calls": 6, "tok_in": 54302, "tok_cached": 38016, "tok_out": 2658, "cost_usd": 0.0, "list_ordinary": 16286, "list_cached": 38016, "list_out": 2658, "list_prompt": 54302, "list_tokens": 56960, "list_token_usd": 0.067528, "list_tool_usd": 0.0, "list_usd": 0.067528, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.7001, "tok_per_call": 9050, "verify_tail": "... [100%]\n3 passed in 0.04s\nVEXIT:0", "pass_": true}
{"task": "extract-elf", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/extract-elf:20251031", "wall_s": 106.7, "agent_exit": 0, "agent_tail": "ase 314ms] credentials/model\ncalling grok-4.6\n\u203a\n [main] \u21af 1 failed\n\u203a\n\u203a\n\u203a\n\u203a\n\u203a\n [main] extract.js parses the ELF and dumps 4-byte little-endian words from .text/.data/.rodata using each section\u2019s virtual address (not a PIE base). Run: `node extract.js /app/a.out > out.json`. On this binary that yields 73 address\u2192int pairs (e.g. 4192 \u2192 4196274163 at .text).\n[usage] 6 api call(s) \u00b7 66409 in (45568 cached, 0 cache writes) + 7484 out tokens \u00b7 $0.0000 \u00b7 6 subscription call(s), flat-rate (not in $)\n", "calls": 6, "tok_in": 66409, "tok_cached": 45568, "tok_out": 7484, "cost_usd": 0.0, "list_ordinary": 20841, "list_cached": 45568, "list_out": 7484, "list_prompt": 66409, "list_tokens": 73893, "list_token_usd": 0.10937, "list_tool_usd": 0.0, "list_usd": 0.10937, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.6862, "tok_per_call": 11068, "verify_tail": ".. [100%]\n2 passed in 1.89s\nVEXIT:0", "pass_": true}
{"task": "kv-store-grpc", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/kv-store-grpc:20251031", "wall_s": 74.5, "agent_exit": 0, "agent_tail": "d, and serving on port 5328.\n\n- grpcio 1.73.0 and grpcio-tools 1.73.0 installed system-wide\n- /app/kv-store.proto defines KVStore with GetVal/SetVal\n- Generated /app/kv_store_pb2.py and /app/kv_store_pb2_grpc.py\n- /app/server.py implements Server with an in-memory dict\n- server.py is running in the background on port 5328 (SetVal/GetVal self-check passed)\n[usage] 14 api call(s) \u00b7 105815 in (71680 cached, 0 cache writes) + 3114 out tokens \u00b7 $0.0000 \u00b7 14 subscription call(s), flat-rate (not in $)\n", "calls": 14, "tok_in": 105815, "tok_cached": 71680, "tok_out": 3114, "cost_usd": 0.0, "list_ordinary": 34135, "list_cached": 71680, "list_out": 3114, "list_prompt": 105815, "list_tokens": 108929, "list_token_usd": 0.122794, "list_tool_usd": 0.0, "list_usd": 0.122794, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.6774, "tok_per_call": 7558, "verify_tail": "....... [100%]\n7 passed in 0.43s\nVEXIT:0", "pass_": true}
{"task": "gcode-to-text", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/gcode-to-text:20251031", "pass_": false, "error": "timeout 900.0s", "wall_s": 1020.4}
{"task": "largest-eigenval", "harness": "graff", "model": "grok-4.6", "image": "alexgshaw/largest-eigenval:20251031", "wall_s": 1008.0, "agent_exit": 0, "agent_tail": "_eigenvalue_and_eigenvector` in `eigen.py` as a small CPython extension (Hessenberg + Francis double-shift QR + eigenvector from real Schur, compiled at import).\n\nOn `eval.py` (median of 100 calls) it is well under the numpy `eig` reference, e.g. ~5\u00b5s vs ~163\u00b5s (2\u00d72) and ~152\u00b5s vs ~452\u00b5s (10\u00d710), and `Ax = \u03bbx` holds for real and complex dominant pairs.\n[usage] 34 api call(s) \u00b7 1248408 in (981120 cached, 0 cache writes) + 50536 out tokens \u00b7 $0.0000 \u00b7 34 subscription call(s), flat-rate (not in $)\n", "calls": 34, "tok_in": 1248408, "tok_cached": 981120, "tok_out": 50536, "cost_usd": 0.0, "list_ordinary": 267288, "list_cached": 981120, "list_out": 50536, "list_prompt": 1248408, "list_tokens": 1298944, "list_token_usd": 2.656704, "list_tool_usd": 0.0, "list_usd": 1.328352, "list_price_kind": "xai-list", "list_high_band": false, "cache_hit": 0.7859, "tok_per_call": 36718, "verify_tail": "........................... [100%]\n27 passed in 1.27s\nVEXIT:0", "pass_": true}