SkillsBench · insights/SKILLS_TASKS_MAP.md source

SkillsBench: skills ↔ tasks map (87 tasks)

Derived from insights/SKILLSBENCH_INVENTORY.md's per-task "Shipped skills" data (the task-by-task-87 run seeds each task from this same per-task skill set — see "Implications for a task-by-task optimization strategy" §1 in that doc).

Table A — all 87 tasks, by category, with skill numbers

software-engineering (16 tasks)

# Task Difficulty Skills (by #)
1 azure-bgp-oscillation-route-leak medium 1
2 data-to-d3 medium 2
3 debug-trl-grpo hard 3, 4, 5
4 dialogue-parser easy 6
5 fix-build-agentops easy 7, 8, 9, 10
6 fix-build-google-auto easy 11, 12, 13
7 fix-visual-stability hard 14, 15, 16
8 flink-query hard 17, 18
9 jax-computing-basics medium 19
10 llm-prefix-cache-replay medium 20, 21
11 parallel-tfidf-search medium 22, 23, 24
12 python-scala-translation medium 25, 26, 27, 28, 29, 30
13 react-performance-debugging hard 14, 15
14 simpo-code-reproduction hard 31, 17
15 spring-boot-jakarta-migration hard 32, 33, 34, 35, 36
16 tictoc-unnecessary-abort-detection hard 37, 38, 39

industrial-physical-systems (14 tasks)

# Task Difficulty Skills (by #)
17 3d-scan-calc hard 40
18 ada-bathroom-plan-repair hard 41, 42, 43
19 adaptive-cruise-control medium 44, 45, 46, 47, 48
20 drone-planning-control medium 49, 50, 51, 52, 53, 54
21 dynamic-object-aware-egomotion medium 55, 56, 57, 58
22 energy-ac-optimal-power-flow medium 59, 60, 61
23 energy-market-pricing hard 62, 63, 64, 61
24 energy-unit-commitment hard 65, 66, 67
25 grid-dispatch-operator medium 62, 63, 61
26 hvac-control medium 68, 69, 70, 71, 72
27 manufacturing-codebook-normalization medium 73, B
28 manufacturing-equipment-maintenance medium B, 74, 75
29 manufacturing-fjsp-optimization medium 76, B
30 r2r-mpc-control medium 77, 78, 79, 80

office-white-collar (14 tasks)

# Task Difficulty Skills (by #)
31 citation-check medium 81
32 court-form-filling easy 17
33 edit-pdf medium 82, 83
34 enterprise-information-search hard 84
35 exceltable-in-ppt medium 85, 86
36 jpg-ocr-stat hard 87, 88, 17, 89, 86
37 latex-formula-extraction medium 90, 17
38 offer-letter-generator easy 91
39 organize-messy-files medium 91, 92, 17, 93, 85
40 paper-anonymizer medium 94, 17
41 pdf-excel-diff medium 17, 86
42 powerlifting-coef-calc easy 95, 96, 86
43 pptx-reference-formatting medium 85
44 sales-pivot-analysis medium 17, 86

natural-science (14 tasks)

# Task Difficulty Skills (by #)
45 crystallographic-wyckoff-position-analysis medium 97, 98
46 earthquake-phase-association hard 99, 100, 101, 102, 103
47 earthquake-plate-calculation medium 104
48 exoplanet-detection-period medium 105, 106, 107, 108, 109
49 flood-risk-analysis medium 110, 111, 112
50 glm-lake-mendota hard 113, 114, 115
51 gravitational-wave-detection medium 116, 117
52 lab-unit-harmonization medium 118
53 lake-warming-attribution medium 119, 120, 121, 122
54 mars-clouds-clustering hard 123, 124, 125
55 protein-expression-analysis medium 86
56 quantum-numerical-simulation medium 126
57 radar-vital-signs medium 127, 128, 129
58 seismic-phase-picking hard 100, 101, 130, 102, 103

finance-economics (9 tasks)

# Task Difficulty Skills (by #)
59 econ-detrending-correlation medium 131
60 financial-modeling-qa hard 17, 86
61 invoice-fraud-detection hard 132, 17, 86
62 reserves-at-risk-calc medium 86
63 sec-financial-report hard 133, 134
64 shock-analysis-demand medium 86
65 shock-analysis-supply hard 86
66 weighted-gdp-calc medium 86
67 xlsx-recover-data medium 135, 86

mathematics-or-formal-reasoning (8 tasks)

# Task Difficulty Skills (by #)
68 bike-rebalance medium 136, 137, 138, 139
69 civ6-adjacency-optimizer hard 140, 141, 142, 143
70 exam-block-sequencing hard 144, 145
71 lean4-proof medium B, B, B, B, 146, 147
72 paratransit-routing hard 148, 149
73 pddl-airport-planning medium 150
74 pddl-tpp-planning medium 150
75 travel-planning medium 151, 152, 153, 154, 155, 156

cybersecurity (7 tasks)

# Task Difficulty Skills (by #)
76 dapt-intrusion-detection hard 157, 158
77 fix-druid-loophole-cve hard 159, 160
78 fix-erlang-ssh-cve hard 161, 162, 163, 164, 165, 166
79 setup-fuzzing-py medium 167, 168, 169
80 software-dependency-audit medium 170, 171, 172
81 suricata-custom-exfil medium 173, 174, 175
82 syzkaller-ppdev-syzlang medium 176, 177, 178

media-content-production (5 tasks)

# Task Difficulty Skills (by #)
83 mario-coin-counting medium 179, 180, 181
84 multilingual-video-dubbing medium 182, 183, 184, 185, 186, 187
85 threejs-structure-parser medium 188, 189
86 threejs-to-obj medium 188, 189
87 video-silence-remover hard 190, 191, 192, 193, 194, 195, 196

Table B — skills index, with the tasks that use each

Rows = skills, numbered by first appearance in Table A. "Categories" lists every category the skill appears in (most skills are single-category; a few — e.g. pdf, xlsx — cross several).

# Skill # tasks Categories Task #s
1 azure-bgp 1 software-engineering 1
2 d3-visualization 1 software-engineering 2
3 grpo 1 software-engineering 3
4 rl-post-training 1 software-engineering 3
5 trl 1 software-engineering 3
6 dialogue-graph 1 software-engineering 4
7 analyze-ci 1 software-engineering 5
8 temporal-python-testing 1 software-engineering 5
9 testing-python 1 software-engineering 5
10 uv-package-manager 1 software-engineering 5
11 maven-build-lifecycle 1 software-engineering 6
12 maven-dependency-management 1 software-engineering 6
13 maven-plugin-configuration 1 software-engineering 6
14 browser-testing 2 software-engineering 7, 13
15 react-best-practices 2 software-engineering 7, 13
16 web-interface-guidelines 1 software-engineering 7
17 pdf 11 finance-economics, office-white-collar, software-engineering 8, 14, 32, 36, 37, 39, 40, 41, 44, 60, 61
18 senior-data-engineer 1 software-engineering 8
19 jax-skills 1 software-engineering 9
20 cache-policy-comparison 1 software-engineering 10
21 prefix-cache-replay 1 software-engineering 10
22 memory-optimization 1 software-engineering 11
23 python-parallelization 1 software-engineering 11
24 workload-balancing 1 software-engineering 11
25 python-scala-collections 1 software-engineering 12
26 python-scala-functional 1 software-engineering 12
27 python-scala-idioms 1 software-engineering 12
28 python-scala-libraries 1 software-engineering 12
29 python-scala-oop 1 software-engineering 12
30 python-scala-syntax-mapping 1 software-engineering 12
31 nlp-research-repo-package-installment 1 software-engineering 14
32 hibernate-upgrade 1 software-engineering 15
33 jakarta-namespace 1 software-engineering 15
34 restclient-migration 1 software-engineering 15
35 spring-boot-migration 1 software-engineering 15
36 spring-security-6 1 software-engineering 15
37 transaction-concurrency-control-foundations 1 software-engineering 16
38 transaction-protocol-reasoning 1 software-engineering 16
39 transaction-trace-analysis 1 software-engineering 16
40 mesh-analysis 1 industrial-physical-systems 17
41 ada-plan-view-accessibility 1 industrial-physical-systems 18
42 architectural-dxf-extraction 1 industrial-physical-systems 18
43 geometric-layout-repair 1 industrial-physical-systems 18
44 csv-processing 1 industrial-physical-systems 19
45 pid-controller 1 industrial-physical-systems 19
46 simulation-metrics 1 industrial-physical-systems 19
47 vehicle-dynamics 1 industrial-physical-systems 19
48 yaml-config 1 industrial-physical-systems 19
49 attitude-controller-planner 1 industrial-physical-systems 20
50 flight-plan-parser 1 industrial-physical-systems 20
51 motor-model-dynamics 1 industrial-physical-systems 20
52 plot-quadrotor 1 industrial-physical-systems 20
53 position-controller-trajectory-planner 1 industrial-physical-systems 20
54 stepinfo-3d 1 industrial-physical-systems 20
55 dyn-object-masks 1 industrial-physical-systems 21
56 egomotion-estimation 1 industrial-physical-systems 21
57 output-validation 1 industrial-physical-systems 21
58 sampling-and-indexing 1 industrial-physical-systems 21
59 ac-branch-pi-model 1 industrial-physical-systems 22
60 casadi-ipopt-nlp 1 industrial-physical-systems 22
61 power-flow-data 3 industrial-physical-systems 22, 23, 25
62 dc-power-flow 2 industrial-physical-systems 23, 25
63 economic-dispatch 2 industrial-physical-systems 23, 25
64 locational-marginal-prices 1 industrial-physical-systems 23
65 milp-solver-workflow 1 industrial-physical-systems 24
66 unit-commitment-data-modeling 1 industrial-physical-systems 24
67 unit-commitment-operating-rules 1 industrial-physical-systems 24
68 excitation-signal-design 1 industrial-physical-systems 26
69 first-order-model-fitting 1 industrial-physical-systems 26
70 imc-tuning-rules 1 industrial-physical-systems 26
71 safety-interlocks 1 industrial-physical-systems 26
72 scipy-curve-fit 1 industrial-physical-systems 26
73 manufacturing-failure-reason-codebook-normalization 1 industrial-physical-systems 27
74 reflow-machine-maintenance-guidance 1 industrial-physical-systems 28
75 reflow-profile-compliance-toolkit 1 industrial-physical-systems 28
76 fjsp-baseline-repair-with-downtime-and-policy 1 industrial-physical-systems 29
77 finite-horizon-lqr 1 industrial-physical-systems 30
78 integral-action-design 1 industrial-physical-systems 30
79 mpc-horizon-tuning 1 industrial-physical-systems 30
80 state-space-linearization 1 industrial-physical-systems 30
81 citation-management 1 office-white-collar 31
82 pdf-editing 1 office-white-collar 33
83 text-parser 1 office-white-collar 33
84 enterprise-artifact-search 1 office-white-collar 34
85 pptx 3 office-white-collar 35, 39, 43
86 xlsx 13 finance-economics, natural-science, office-white-collar 35, 36, 41, 42, 44, 55, 60, 61, 62, 64, 65, 66, 67
87 image-ocr 1 office-white-collar 36
88 openai-vision 1 office-white-collar 36
89 video-frame-extraction 1 office-white-collar 36
90 marker 1 office-white-collar 37
91 docx 2 office-white-collar 38, 39
92 file-organizer 1 office-white-collar 39
93 planning-with-files 1 office-white-collar 39
94 academic-pdf-redaction 1 office-white-collar 40
95 powerlifting 1 office-white-collar 42
96 senior-data-scientist 1 office-white-collar 42
97 pymatgen 1 natural-science 45
98 sympy 1 natural-science 45
99 gamma-phase-associator 1 natural-science 46
100 licenses 2 natural-science 46, 58
101 obspy-data-api 2 natural-science 46, 58
102 seisbench-model-api 2 natural-science 46, 58
103 seismic-picker-selection 2 natural-science 46, 58
104 geospatial-analysis 1 natural-science 47
105 box-least-squares 1 natural-science 48
106 exoplanet-workflows 1 natural-science 48
107 light-curve-preprocessing 1 natural-science 48
108 lomb-scargle-periodogram 1 natural-science 48
109 transit-least-squares 1 natural-science 48
110 flood-detection 1 natural-science 49
111 nws-flood-thresholds 1 natural-science 49
112 usgs-data-download 1 natural-science 49
113 glm-basics 1 natural-science 50
114 glm-calibration 1 natural-science 50
115 glm-output 1 natural-science 50
116 conditioning 1 natural-science 51
117 matched-filtering 1 natural-science 51
118 lab-unit-harmonization 1 natural-science 52
119 contribution-analysis 1 natural-science 53
120 meteorology-driver-classification 1 natural-science 53
121 pca-decomposition 1 natural-science 53
122 trend-analysis 1 natural-science 53
123 custom-distance-metrics 1 natural-science 54
124 parallel-processing 1 natural-science 54
125 pareto-optimization 1 natural-science 54
126 qutip 1 natural-science 56
127 radar-signal-processing 1 natural-science 57
128 radar-vital-signs 1 natural-science 57
129 vital-sign-extraction 1 natural-science 57
130 obspy-datacenter-client 1 natural-science 58
131 timeseries-detrending 1 finance-economics 59
132 fuzzy-match 1 finance-economics 61
133 13f-analyzer 1 finance-economics 63
134 fuzzy-name-search 1 finance-economics 63
135 data-reconciliation 1 finance-economics 67
136 geospatial-routing-data 1 mathematics-or-formal-reasoning 68
137 logistics-rules-to-optimization 1 mathematics-or-formal-reasoning 68
138 routing-subtour-elimination 1 mathematics-or-formal-reasoning 68
139 scip-opt 1 mathematics-or-formal-reasoning 68
140 civ6lib 1 mathematics-or-formal-reasoning 69
141 hex-grid-spatial 1 mathematics-or-formal-reasoning 69
142 map-optimization-strategy 1 mathematics-or-formal-reasoning 69
143 sqlite-map-parser 1 mathematics-or-formal-reasoning 69
144 mip-solver-and-solution-audit 1 mathematics-or-formal-reasoning 70
145 ordered-window-sequencing-mip 1 mathematics-or-formal-reasoning 70
146 lean4-memories 1 mathematics-or-formal-reasoning 71
147 lean4-theorem-proving 1 mathematics-or-formal-reasoning 71
148 ortools-pickup-delivery-routing 1 mathematics-or-formal-reasoning 72
149 ortools-routing-modeling 1 mathematics-or-formal-reasoning 72
150 pddl-skills 2 mathematics-or-formal-reasoning 73, 74
151 search-accommodations 1 mathematics-or-formal-reasoning 75
152 search-attractions 1 mathematics-or-formal-reasoning 75
153 search-cities 1 mathematics-or-formal-reasoning 75
154 search-driving-distance 1 mathematics-or-formal-reasoning 75
155 search-flights 1 mathematics-or-formal-reasoning 75
156 search-restaurants 1 mathematics-or-formal-reasoning 75
157 pcap-analysis 1 cybersecurity 76
158 threat-detection 1 cybersecurity 76
159 jackson-security 1 cybersecurity 77
160 senior-java 1 cybersecurity 77
161 erlang-concurrency 1 cybersecurity 78
162 erlang-distribution 1 cybersecurity 78
163 erlang-otp-behaviors 1 cybersecurity 78
164 find-bugs 1 cybersecurity 78
165 senior-security 1 cybersecurity 78
166 ssh-penetration-testing 1 cybersecurity 78
167 discover-important-function 1 cybersecurity 79
168 fuzzing-python 1 cybersecurity 79
169 setup-env 1 cybersecurity 79
170 cvss-score-extraction 1 cybersecurity 80
171 trivy-offline-vulnerability-scanning 1 cybersecurity 80
172 vulnerability-csv-reporting 1 cybersecurity 80
173 pcap-triage-tshark 1 cybersecurity 81
174 suricata-offline-evejson 1 cybersecurity 81
175 suricata-rules-basics 1 cybersecurity 81
176 syz-extract-constants 1 cybersecurity 82
177 syzkaller-build-loop 1 cybersecurity 82
178 syzlang-ioctl-basics 1 cybersecurity 82
179 ffmpeg-keyframe-extraction 1 media-content-production 83
180 image-editing 1 media-content-production 83
181 object-counter 1 media-content-production 83
182 ffmpeg-audio-processing 1 media-content-production 84
183 ffmpeg-format-conversion 1 media-content-production 84
184 ffmpeg-media-info 1 media-content-production 84
185 ffmpeg-video-editing 1 media-content-production 84
186 ffmpeg-video-filters 1 media-content-production 84
187 text-to-speech 1 media-content-production 84
188 obj-exporter 2 media-content-production 85, 86
189 threejs 2 media-content-production 85, 86
190 audio-extractor 1 media-content-production 87
191 energy-calculator 1 media-content-production 87
192 pause-detector 1 media-content-production 87
193 report-generator 1 media-content-production 87
194 segment-combiner 1 media-content-production 87
195 silence-detector 1 media-content-production 87
196 video-processor 1 media-content-production 87

Table C — per-category train/test split (skill-disjoint-safe)

For each category: a test subset of tasks whose skills all also appear in at least one train task of the same category — i.e. every skill the test set needs to succeed is still present, unseen-combination, in the train set. Intended use: optimize/tune skills on the train tasks, then evaluate held-out on the test tasks. Target was ~20–30% of each category, found greedily (tasks with the fewest skills tried first, since they're least likely to strand a skill). Where a skill is unique to a single task within its category, that task can never be a test task — this is what caps the achievable split, sometimes to 0%.

Category n Test tasks (n, %) Train task #s Test task #s
software-engineering 16 1 (6.2%) 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 14, 15, 16 13
industrial-physical-systems 14 1 (7.1%) 17, 18, 19, 20, 21, 22, 23, 24, 26, 27, 28, 29, 30 25
office-white-collar 14 4 (28.6%) 31, 33, 34, 36, 37, 39, 40, 41, 42, 44 32, 35, 38, 43
natural-science 14 0 (0.0%) 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58 —
finance-economics 9 3 (33.3%) 59, 60, 61, 63, 66, 67 62, 64, 65
mathematics-or-formal-reasoning 8 1 (12.5%) 68, 69, 70, 71, 72, 74, 75 73
cybersecurity 7 0 (0.0%) 76, 77, 78, 79, 80, 81, 82 —
media-content-production 5 1 (20.0%) 83, 84, 86, 87 85

Table C detail — test task names by category

software-engineering — 1/16 (6.2%) - react-performance-debugging (#13, skills 14, 15)

industrial-physical-systems — 1/14 (7.1%) - grid-dispatch-operator (#25, skills 62, 63, 61)

office-white-collar — 4/14 (28.6%) - court-form-filling (#32, skills 17) - exceltable-in-ppt (#35, skills 85, 86) - offer-letter-generator (#38, skills 91) - pptx-reference-formatting (#43, skills 85)

natural-science — 0/14 (0.0%) - No valid split found — every task in this category has at least one skill unique to it within the category.

finance-economics — 3/9 (33.3%) - reserves-at-risk-calc (#62, skills 86) - shock-analysis-demand (#64, skills 86) - shock-analysis-supply (#65, skills 86)

mathematics-or-formal-reasoning — 1/8 (12.5%) - pddl-airport-planning (#73, skills 150)

cybersecurity — 0/7 (0.0%) - No valid split found — every task in this category has at least one skill unique to it within the category.

media-content-production — 1/5 (20.0%) - threejs-structure-parser (#85, skills 188, 189)

Boilerplate entries (not numbered as skills)

Filename # tasks Task #s
reference.md 3 27, 28, 29
INSTALLATION.md 1 71
LICENSE 1 71
README.md 1 71
TESTING.md 1 71

Capability-change classification (Tables D and E)

For each of the 87 tasks, ui/heatmap.html's DATA array names the accepted best candidate (best_tag, "seed" if the optimizer never improved on seed). For every task where best_tag != "seed" (52 tasks), this walks that task's run directory across the six c1-c5/v2 worktrees, diffs candidates/seed/<skill>/ against candidates/<best_tag>/<skill>/ file-by-file (excluding cap-evolve's own INSTRUCTIONS.md/PROCESS.md), and classifies every changed file per skill. Row key is (task, skill), not (task, best-candidate) alone: a task has exactly one accepted candidate, but that one candidate snapshot can touch several skill packages at once (e.g. energy-ac-optimal-power-flow's cand_0003 touches both ac-branch-pi-model and casadi-ipopt-nlp), so a single task-level row would collapse independent per-skill changes into one ambiguous line. Best-candidate ID and its val score are metadata columns on each row, constant across all of a task's rows.

Disambiguation: 14 of the 52 changed tasks have more than one run_task_<id>* directory across the six worktrees (reruns, corrupted batches, infra-broken attempts). Each run's state.json.best_id is compared against heatmap.html's recorded best_tag for that task; a unique match is used. Where more than one run matches (energy-ac-optimal-power-flow: both its v2 and its v1_CORRUPTED_batch3 run report the same best_id), the non-corrupted run is preferred — for that task this agrees with heatmap.html's own source: "c2-v2" field, i.e. it independently confirms the corrupted run is correctly excluded. This script does not attempt to adjudicate prev-vs-curr for the 7 tasks with a DATA_C4 rerun entry (crystallographic-wyckoff-position-analysis, energy-market-pricing, fix-erlang-ssh-cve, flink-query, invoice-fraud-detection, organize-messy-files, shock-analysis-demand) — it takes whatever the main DATA array already records as authoritative, since several of those reruns are regressions or fall back to best_tag: "seed" rather than uniformly superseding the original run, so "latest wins" is not a safe blanket rule there. That reconciliation is a separate, open question from this classification and is not resolved here.

Columns: skill-added/skill-deleted — the whole top-level skill directory is new/missing relative to seed (blocked once fixed_vocabulary: true, per Step 2, ships). skill-prompt — SKILL.md content changed (0/1). skill-package — count of added/removed/edited non-SKILL.md, non-script package files (references/*.md, forms.md, etc.). tool-added/tool-deleted/ tool-edited — counts of added/removed/edited files under scripts/ or a top-level executable script (e.g. recalc.py). other — anything unclassified (empty in this data).

One task, fix-erlang-ssh-cve, produced zero package-level file changes despite its accepted candidate scoring above seed (0.6 -> 0.9) — the improvement is not attributable to any skill edit in this diff; see Unresolved below.

Table D — per-(task, skill) capability changes, best candidate vs seed

task skill best_candidate val skill-added skill-deleted skill-prompt skill-package tool-added tool-deleted tool-edited other
ada-bathroom-plan-repair ada-plan-view-accessibility cand_0003 1.0 0 0 1 0 0 0 0 0
ada-bathroom-plan-repair architectural-dxf-extraction cand_0003 1.0 0 0 1 0 1 0 0 0
ada-bathroom-plan-repair geometric-layout-repair cand_0003 1.0 0 0 1 0 2 0 0 0
adaptive-cruise-control pid-controller cand_0003 1.0 0 0 1 0 2 0 0 0
adaptive-cruise-control simulation-metrics cand_0003 1.0 0 0 1 0 1 0 0 0
adaptive-cruise-control vehicle-dynamics cand_0003 1.0 0 0 1 0 2 0 0 0
azure-bgp-oscillation-route-leak azure-bgp cand_0001 1.0 0 0 1 0 0 0 0 0
bike-rebalance logistics-rules-to-optimization cand_0002 1.0 0 0 1 0 1 0 0 0
bike-rebalance scip-opt cand_0002 1.0 0 0 1 0 0 0 0 0
civ6-adjacency-optimizer civ6lib cand_0002 1.0 0 0 1 0 3 0 0 0
court-form-filling pdf cand_0002 1.0 0 0 0 1 0 0 0 0
crystallographic-wyckoff-position-analysis pymatgen cand_0002 0.835 0 0 1 0 0 0 0 0
data-to-d3 d3-visualization cand_0002 1.0 0 0 1 0 0 0 1 0
debug-trl-grpo grpo cand_0004 1.0 0 0 1 0 0 0 0 0
debug-trl-grpo rl-post-training cand_0004 1.0 0 0 1 0 0 0 1 0
debug-trl-grpo trl cand_0004 1.0 0 0 1 0 0 0 0 0
dynamic-object-aware-egomotion dyn-object-masks cand_0001 1.0 0 0 1 0 1 0 0 0
dynamic-object-aware-egomotion egomotion-estimation cand_0001 1.0 0 0 1 0 1 0 0 0
earthquake-plate-calculation geospatial-analysis cand_0001 1.0 0 0 1 0 0 0 0 0
edit-pdf pdf-editing cand_0002 1.0 0 0 1 0 1 0 0 0
energy-ac-optimal-power-flow ac-branch-pi-model cand_0003 1.0 0 0 1 0 1 0 0 0
energy-ac-optimal-power-flow casadi-ipopt-nlp cand_0003 1.0 0 0 1 0 1 0 0 0
energy-market-pricing dc-power-flow cand_0002 0.9 0 0 1 0 0 0 0 0
energy-market-pricing economic-dispatch cand_0002 0.9 0 0 1 0 0 0 0 0
energy-market-pricing locational-marginal-prices cand_0002 0.9 0 0 1 0 0 0 0 0
energy-market-pricing power-flow-data cand_0002 0.9 0 0 1 0 1 0 0 0
energy-unit-commitment milp-solver-workflow cand_0002 1.0 0 0 1 0 0 0 0 0
energy-unit-commitment unit-commitment-operating-rules cand_0002 1.0 0 0 1 0 1 0 0 0
enterprise-information-search enterprise-artifact-search cand_0001 1.0 0 0 1 0 1 0 0 0
exam-block-sequencing mip-solver-and-solution-audit cand_0002 1.0 0 0 1 0 0 0 0 0
exam-block-sequencing ordered-window-sequencing-mip cand_0002 1.0 0 0 1 0 1 0 0 0
exceltable-in-ppt pptx cand_0001 1.0 0 0 1 0 1 0 0 0
exceltable-in-ppt xlsx cand_0001 1.0 0 0 1 0 0 0 0 0
flink-query senior-data-engineer cand_0004 0.9 0 0 1 1 0 0 0 0
grid-dispatch-operator dc-power-flow cand_0002 1.0 0 0 1 0 0 0 0 0
grid-dispatch-operator economic-dispatch cand_0002 1.0 0 0 1 0 2 0 0 0
grid-dispatch-operator power-flow-data cand_0002 1.0 0 0 1 0 0 0 0 0
invoice-fraud-detection fuzzy-match cand_0002 0.4 0 0 1 0 0 0 0 0
jpg-ocr-stat image-ocr cand_0001 1.0 0 0 1 0 1 0 0 0
lab-unit-harmonization lab-unit-harmonization cand_0001 1.0 0 0 1 0 1 0 0 0
lake-warming-attribution contribution-analysis cand_0002 1.0 0 0 1 0 1 0 0 0
lake-warming-attribution meteorology-driver-classification cand_0002 1.0 0 0 1 0 0 0 0 0
lake-warming-attribution trend-analysis cand_0002 1.0 0 0 1 0 1 0 0 0
latex-formula-extraction marker cand_0004 0.8 0 0 1 0 1 0 0 0
lean4-proof lean4-theorem-proving cand_0001 1.0 0 0 1 0 0 0 0 0
manufacturing-codebook-normalization manufacturing-failure-reason-codebook-normalization cand_0002 1.0 0 0 1 0 1 0 0 0
manufacturing-equipment-maintenance reflow-machine-maintenance-guidance cand_0003 1.0 0 0 1 0 1 0 0 0
manufacturing-equipment-maintenance reflow-profile-compliance-toolkit cand_0003 1.0 0 0 1 0 1 0 0 0
manufacturing-fjsp-optimization fjsp-baseline-repair-with-downtime-and-policy cand_0001 1.0 0 0 1 0 1 0 0 0
multilingual-video-dubbing text-to-speech cand_0004 1.0 0 0 1 0 1 0 0 0
organize-messy-files file-organizer cand_0003 0.9 0 0 1 0 0 0 0 0
paratransit-routing ortools-pickup-delivery-routing cand_0001 1.0 0 0 1 0 1 0 0 0
paratransit-routing ortools-routing-modeling cand_0001 1.0 0 0 1 0 0 0 0 0
pddl-airport-planning pddl-skills cand_0002 1.0 0 0 1 0 2 0 0 0
pdf-excel-diff pdf cand_0001 1.0 0 0 1 0 1 0 0 0
r2r-mpc-control finite-horizon-lqr cand_0001 1.0 0 0 1 0 1 0 0 0
r2r-mpc-control mpc-horizon-tuning cand_0001 1.0 0 0 1 0 0 0 0 0
react-performance-debugging react-best-practices cand_0001 1.0 0 0 1 2 1 0 0 0
reserves-at-risk-calc xlsx cand_0004 1.0 0 0 1 0 1 0 0 0
sales-pivot-analysis xlsx cand_0001 1.0 0 0 1 0 0 0 0 0
sec-financial-report 13f-analyzer cand_0002 1.0 0 0 1 0 0 0 1 0
setup-fuzzing-py discover-important-function cand_0004 0.549 0 0 1 0 0 0 0 0
setup-fuzzing-py fuzzing-python cand_0004 0.549 0 0 1 0 0 0 0 0
shock-analysis-demand xlsx cand_0004 0.9 0 0 1 1 1 0 0 0
shock-analysis-supply xlsx cand_0004 0.3 0 0 1 1 1 0 0 0
simpo-code-reproduction nlp-research-repo-package-installment cand_0002 1.0 0 0 1 0 1 0 0 0
software-dependency-audit trivy-offline-vulnerability-scanning cand_0001 1.0 0 0 1 0 1 0 0 0
software-dependency-audit vulnerability-csv-reporting cand_0001 1.0 0 0 1 0 0 0 0 0
syzkaller-ppdev-syzlang syz-extract-constants cand_0002 1.0 0 0 1 0 1 0 0 0
syzkaller-ppdev-syzlang syzkaller-build-loop cand_0002 1.0 0 0 1 0 0 0 0 0
syzkaller-ppdev-syzlang syzlang-ioctl-basics cand_0002 1.0 0 0 1 0 0 0 0 0
threejs-structure-parser threejs cand_0001 1.0 0 0 1 0 1 0 1 0
tictoc-unnecessary-abort-detection transaction-trace-analysis cand_0001 1.0 0 0 1 1 1 0 0 0
travel-planning search-attractions cand_0003 1.0 0 0 1 0 1 0 1 0
video-silence-remover pause-detector cand_0002 1.0 0 0 1 0 0 0 1 0
video-silence-remover silence-detector cand_0002 1.0 0 0 1 0 0 0 1 0
video-silence-remover video-processor cand_0002 1.0 0 0 1 0 1 0 0 0
weighted-gdp-calc xlsx cand_0001 1.0 0 0 1 0 1 0 0 0
xlsx-recover-data data-reconciliation cand_0002 1.0 0 0 1 0 1 0 0 0
TOTAL 70 distinct 51 tasks 0 0 78 7 52 0 7 0

Table E — skill-level rollup, cross-task conflicts

skill tasks touched conflict skill-added skill-deleted skill-prompt skill-package tool-added tool-deleted tool-edited other
13f-analyzer 1 0 0 1 0 0 0 1 0
ac-branch-pi-model 1 0 0 1 0 1 0 0 0
ada-plan-view-accessibility 1 0 0 1 0 0 0 0 0
architectural-dxf-extraction 1 0 0 1 0 1 0 0 0
azure-bgp 1 0 0 1 0 0 0 0 0
casadi-ipopt-nlp 1 0 0 1 0 1 0 0 0
civ6lib 1 0 0 1 0 3 0 0 0
contribution-analysis 1 0 0 1 0 1 0 0 0
d3-visualization 1 0 0 1 0 0 0 1 0
data-reconciliation 1 0 0 1 0 1 0 0 0
dc-power-flow 2 yes 0 0 2 0 0 0 0 0
discover-important-function 1 0 0 1 0 0 0 0 0
dyn-object-masks 1 0 0 1 0 1 0 0 0
economic-dispatch 2 yes 0 0 2 0 2 0 0 0
egomotion-estimation 1 0 0 1 0 1 0 0 0
enterprise-artifact-search 1 0 0 1 0 1 0 0 0
file-organizer 1 0 0 1 0 0 0 0 0
finite-horizon-lqr 1 0 0 1 0 1 0 0 0
fjsp-baseline-repair-with-downtime-and-policy 1 0 0 1 0 1 0 0 0
fuzzing-python 1 0 0 1 0 0 0 0 0
fuzzy-match 1 vocab-violation¹ 0 0 1 0 0 0 0 0
geometric-layout-repair 1 0 0 1 0 2 0 0 0
geospatial-analysis 1 0 0 1 0 0 0 0 0
grpo 1 0 0 1 0 0 0 0 0
image-ocr 1 0 0 1 0 1 0 0 0
lab-unit-harmonization 1 0 0 1 0 1 0 0 0
lean4-theorem-proving 1 0 0 1 0 0 0 0 0
locational-marginal-prices 1 0 0 1 0 0 0 0 0
logistics-rules-to-optimization 1 0 0 1 0 1 0 0 0
manufacturing-failure-reason-codebook-normalization 1 0 0 1 0 1 0 0 0
marker 1 0 0 1 0 1 0 0 0
meteorology-driver-classification 1 0 0 1 0 0 0 0 0
milp-solver-workflow 1 0 0 1 0 0 0 0 0
mip-solver-and-solution-audit 1 0 0 1 0 0 0 0 0
mpc-horizon-tuning 1 0 0 1 0 0 0 0 0
nlp-research-repo-package-installment 1 0 0 1 0 1 0 0 0
ordered-window-sequencing-mip 1 0 0 1 0 1 0 0 0
ortools-pickup-delivery-routing 1 0 0 1 0 1 0 0 0
ortools-routing-modeling 1 0 0 1 0 0 0 0 0
pause-detector 1 0 0 1 0 0 0 1 0
pddl-skills 1 0 0 1 0 2 0 0 0
pdf 2 yes 0 0 1 1 1 0 0 0
pdf-editing 1 0 0 1 0 1 0 0 0
pid-controller 1 0 0 1 0 2 0 0 0
power-flow-data 2 yes 0 0 2 0 1 0 0 0
pptx 1 0 0 1 0 1 0 0 0
pymatgen 1 0 0 1 0 0 0 0 0
react-best-practices 1 0 0 1 2 1 0 0 0
reflow-machine-maintenance-guidance 1 0 0 1 0 1 0 0 0
reflow-profile-compliance-toolkit 1 0 0 1 0 1 0 0 0
rl-post-training 1 0 0 1 0 0 0 1 0
scip-opt 1 0 0 1 0 0 0 0 0
search-attractions 1 0 0 1 0 1 0 1 0
senior-data-engineer 1 0 0 1 1 0 0 0 0
silence-detector 1 0 0 1 0 0 0 1 0
simulation-metrics 1 0 0 1 0 1 0 0 0
syz-extract-constants 1 0 0 1 0 1 0 0 0
syzkaller-build-loop 1 0 0 1 0 0 0 0 0
syzlang-ioctl-basics 1 0 0 1 0 0 0 0 0
text-to-speech 1 0 0 1 0 1 0 0 0
threejs 1 0 0 1 0 1 0 1 0
transaction-trace-analysis 1 0 0 1 1 1 0 0 0
trend-analysis 1 0 0 1 0 1 0 0 0
trivy-offline-vulnerability-scanning 1 0 0 1 0 1 0 0 0
trl 1 0 0 1 0 0 0 0 0
unit-commitment-operating-rules 1 0 0 1 0 1 0 0 0
vehicle-dynamics 1 0 0 1 0 2 0 0 0
video-processor 1 0 0 1 0 1 0 0 0
vulnerability-csv-reporting 1 0 0 1 0 0 0 0 0
xlsx 6 yes 0 0 6 2 4 0 0 0
TOTAL 70 skills 6 conflicted 0 0 78 7 52 0 7 0

¹ fuzzy-match is not a clean cross-task conflict under the DATA-authoritative source used here (its seed only serves invoice-fraud-detection); it is counted with the conflict set because a separate, since-excluded candidate for energy-ac-optimal-power-flow (the v1_CORRUPTED_batch3 run, superseded by v2 in heatmap.html's own source field) invented an out-of-vocabulary fuzzy-match package — see evidence/energy-ac-optimal-power-flow-vocabulary-violation/. This is the 6th skill in the plan's original "6 skills edited under 2+ tasks" count: 5 clean conflicts (dc-power-flow, economic-dispatch, pdf, power-flow-data, xlsx) plus this one vocabulary-violation case.

Provenance — conflicted / vocab-violation skills

dc-power-flow - energy-market-pricing -> intake_skillbench_c2/.capevolve/run_task_energy-market-pricing_v2 / cand_0002 (val 0.9) - grid-dispatch-operator -> intake_skillbench_c2/.capevolve/run_task_grid-dispatch-operator_v1 / cand_0002 (val 1.0)

economic-dispatch - energy-market-pricing -> intake_skillbench_c2/.capevolve/run_task_energy-market-pricing_v2 / cand_0002 (val 0.9) - grid-dispatch-operator -> intake_skillbench_c2/.capevolve/run_task_grid-dispatch-operator_v1 / cand_0002 (val 1.0)

fuzzy-match - invoice-fraud-detection -> intake_skillbench_c2/.capevolve/run_task_invoice-fraud-detection_v1 / cand_0002 (val 0.4) - energy-ac-optimal-power-flow (excluded) -> intake_skillbench_c2/.capevolve/run_task_energy-ac-optimal-power-flow_v1_CORRUPTED_batch3 / cand_0002/cand_0003 — out-of-vocabulary, superseded by v2 in heatmap.html

pdf - court-form-filling -> intake_skillbench_c2/.capevolve/run_task_court-form-filling_v1 / cand_0002 (val 1.0) - pdf-excel-diff -> intake_skillbench_c2/.capevolve/run_task_pdf-excel-diff_v1_KILLED_ceiling_reached_1.0 / cand_0001 (val 1.0)

power-flow-data - energy-market-pricing -> intake_skillbench_c2/.capevolve/run_task_energy-market-pricing_v2 / cand_0002 (val 0.9) - grid-dispatch-operator -> intake_skillbench_c2/.capevolve/run_task_grid-dispatch-operator_v1 / cand_0002 (val 1.0)

xlsx - exceltable-in-ppt -> intake_skillbench_c2/.capevolve/run_task_exceltable-in-ppt_v1_KILLED_ceiling_reached_1.0 / cand_0001 (val 1.0) - reserves-at-risk-calc -> intake_skillbench_c2/.capevolve/run_task_reserves-at-risk-calc_v2 / cand_0004 (val 1.0) - sales-pivot-analysis -> intake_skillbench_c2/.capevolve/run_task_sales-pivot-analysis_v1_KILLED_ceiling_reached_1.0 / cand_0001 (val 1.0) - shock-analysis-demand -> intake_skillbench_c2/.capevolve/run_task_shock-analysis-demand_v2 / cand_0004 (val 0.9) - shock-analysis-supply -> intake_skillbench_c2/.capevolve/run_task_shock-analysis-supply_v2 / cand_0004 (val 0.3) - weighted-gdp-calc -> intake_skillbench_c2/.capevolve/run_task_weighted-gdp-calc_v1_KILLED_ceiling_reached_1.0 / cand_0001 (val 1.0)

Unresolved