Tool Usage Patterns
analysis of 185,537 assistant messages across 4,259 threads.
Tool Frequency (Overall)
| tool | mentions |
|---|---|
| Bash | 44,681 |
| edit_file | 42,195 |
| Read | 38,019 |
| Grep | 13,991 |
| create_file | 3,630 |
| oracle | 1,279 |
| Task | 911 |
| glob | 824 |
| read_web_page | 763 |
| web_search | 594 |
| finder | 237 |
| librarian | 198 |
the core trio (Bash, edit_file, Read) dominates — these are the workhorses. oracle and Task are used sparingly but strategically.
Tool Combinations Per Thread
pattern format: [Read|Grep|Bash|edit|create|oracle|finder]
| pattern | threads | interpretation |
|---|---|---|
| 1111000 | 798 | read+grep+bash+edit (standard dev flow) |
| 1111100 | 502 | above + create_file |
| 1011000 | 260 | read+bash+edit (no grep) |
| 1111110 | 247 | full stack with oracle |
| 1111010 | 229 | full stack with finder |
observation: most threads use 4+ tools. the “full stack” pattern (all major tools) appears in ~12% of threads.
Tool Usage by Outcome
| status | threads | edit_uses | oracle_uses | bash_uses |
|---|---|---|---|---|
| RESOLVED | 2,070 | 29,480 | 2,467 | 30,053 |
| COMMITTED | 305 | 2,861 | 290 | 4,783 |
| FRUSTRATED | 13 | 300 | 29 | 286 |
| HANDOFF | 573 | 3,653 | 172 | 6,355 |
| UNKNOWN | 1,184 | 5,832 | 450 | 4,668 |
Normalized by Thread Count
| status | avg assistant msgs | avg msg length |
|---|---|---|
| RESOLVED | 59.9 | 759 chars |
| FRUSTRATED | 80.0 | 839 chars |
| STUCK | 117.0 | 748 chars |
| EXPLORATORY | 5.6 | 509 chars |
key insight: FRUSTRATED threads have MORE messages (80 avg) than RESOLVED (60 avg). this suggests frustration comes from thrashing, not lack of effort.
Tool Adoption Rates by Outcome
| status | threads | % oracle | % finder | % librarian | % Task |
|---|---|---|---|---|---|
| RESOLVED | 2,070 | 25.0% | 11.1% | 4.3% | 40.5% |
| FRUSTRATED | 13 | 46.2% | 15.4% | 7.7% | 61.5% |
| COMMITTED | 305 | 22.3% | 12.1% | 3.3% | 34.1% |
counterintuitive: FRUSTRATED threads actually use oracle MORE (46% vs 25%). this doesn’t mean oracle causes frustration — likely users reach for oracle when already stuck.
Tool Mastery Progression Over Time
| month | threads | oracle | finder | librarian | subagent | resolve % |
|---|---|---|---|---|---|---|
| 2025-05 | 24 | 0 | 0 | 0 | 0 | 85.1% |
| 2025-06 | 288 | 0 | 31 | 0 | 50 | 60.8% |
| 2025-07 | 321 | 154 | 130 | 0 | 32 | 79.5% |
| 2025-08 | 281 | 491 | 0 | 0 | 91 | 72.3% |
| 2025-09 | 245 | 631 | 2 | 0 | 66 | 79.9% |
| 2025-10 | 260 | 438 | 67 | 44 | 60 | 81.5% |
| 2025-11 | 416 | 360 | 99 | 71 | 76 | 69.2% |
| 2025-12 | 1,417 | 988 | 223 | 154 | 159 | 73.9% |
| 2026-01 | 1,007 | 354 | 95 | 71 | 102 | 41.5% |
progression signals:
- oracle adoption spikes in jul 2025 (first significant use)
- librarian appears in oct 2025
- resolve rate peaks at 81.5% in oct 2025
- jan 2026 shows low resolve rate (41.5%) — likely incomplete threads
Key Findings
- core workflow is Bash + edit_file + Read — accounts for bulk of tool usage
- more messages ≠ better outcomes — frustrated threads average 33% more messages
- oracle is a “stuck” signal — higher adoption in frustrated threads suggests it’s reached for when things go wrong
- finder is underutilized — only 11% of resolved threads use it
- subagent (Task) correlates with frustration — 61.5% in frustrated vs 40.5% in resolved
- oct 2025 was the “golden month” — highest resolve rate, balanced tool adoption
Recommendations
- investigate why Task usage is higher in frustrated threads — could be over-delegation
- finder adoption remains low; might benefit from better prompting
- oracle as “last resort” pattern is concerning — could be integrated earlier