Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Team-wide charter and history optimization through skill extraction
.claude/skills/github-reskill/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -30% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -28% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -37% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -45% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -23% | 0% |
When the coordinator hears "team, reskill" (or similar: "optimize context", "slim down charters"), trigger a team-wide optimization pass. The goal: reduce per-agent context consumption by extracting shared patterns from charters and histories into reusable skills.
This is a periodic maintenance activity. Run whenever charter/history bloat is suspected.
Read all agent charters and histories. Measure byte sizes. Identify:
For each identified pattern:
.squad/skills/{skill-name}/SKILL.mdCharters — target ≤1.5KB per agent:
Preferred: {model}Histories — target ≤8KB per agent:
## Core Context sectionOutput a savings table:
| Agent | Charter Before | Charter After | History Before | History After | Saved | |-------|---------------|---------------|----------------|---------------|-------|
Include totals and percentage reduction.
# {Name} — {Role}
> {Tagline — one sentence capturing voice and philosophy}
## Identity
- **Name:** {Name}
- **Role:** {Role}
- **Expertise:** {comma-separated list}
## What I Own
- {bullet list of owned artifacts/domains}
## How I Work
- {unique patterns and principles — NOT boilerplate}
## Boundaries
**I handle:** {domain list}
**I don't handle:** {explicit exclusions}
## Model
Preferred: {model}.squad/decisions.md during reskill| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 6,332 | 8,875 | +40% | 1 | 1 | 0% | 279 | 1,436 | +415% | 0 | 0 | — |
case-02 | fail→fail | 4,155 | 8,636 | +108% | 1 | 1 | 0% | 156 | 1,273 | +716% | 0 | 0 | — |
case-03 | fail→fail | 26,304 | 8,234 | -69% | 1 | 1 | 0% | 3,912 | 1,250 | -68% | 0 | 0 | — |
case-04 | fail→pass | 21,969 | 4,178 | -81% | 1 | 1 | 0% | 1,961 | 1,365 | -30% | 0 | 0 | — |
case-05 | fail→pass | 13,092 | 4,526 | -65% | 1 | 1 | 0% | 1,927 | 1,384 | -28% | 0 | 0 | — |
case-06 | fail→pass | 14,551 | 4,699 | -68% | 1 | 1 | 0% | 2,141 | 1,343 | -37% | 0 | 0 | — |
case-07 | pass→pass | 12,172 | 3,759 | -69% | 1 | 1 | 0% | 1,601 | 1,383 | -14% | 0 | 0 | — |
case-08 | pass→fail | 12,066 | 5,801 | -52% | 1 | 1 | 0% | 1,784 | 1,484 | -17% | 0 | 0 | — |
case-09 | fail→pass | 12,845 | 2,513 | -80% | 1 | 1 | 0% | 1,971 | 1,087 | -45% | 0 | 0 | — |
case-10 | fail→pass | 10,974 | 3,038 | -72% | 1 | 1 | 0% | 1,568 | 1,204 | -23% | 0 | 0 | — |
case-11 | fail→pass | 10,019 | 2,875 | -71% | 1 | 1 | 0% | 1,394 | 1,143 | -18% | 0 | 0 | — |
case-12 | fail→pass | 17,101 | 3,405 | -80% | 1 | 1 | 0% | 2,882 | 1,249 | -57% | 0 | 0 | — |
case-13 | fail→fail | 10,466 | 4,848 | -54% | 1 | 1 | 0% | 1,372 | 1,514 | +10% | 0 | 0 | — |
case-14 | fail→pass | 10,088 | 3,918 | -61% | 1 | 1 | 0% | 1,514 | 1,303 | -14% | 0 | 0 | — |
case-15 | fail→pass | 14,322 | 15,895 | +11% | 1 | 1 | 0% | 1,287 | 2,047 | +59% | 0 | 0 | — |
case-16 | fail→pass | 10,317 | 2,847 | -72% | 1 | 1 | 0% | 1,714 | 1,149 | -33% | 0 | 0 | — |
case-17 | fail→pass | 13,111 | 3,040 | -77% | 1 | 1 | 0% | 1,784 | 1,158 | -35% | 0 | 0 | — |
case-18 | fail→pass | 20,534 | 4,056 | -80% | 1 | 1 | 0% | 1,679 | 1,269 | -24% | 0 | 0 | — |
case-19 | fail→fail | 11,885 | 3,762 | -68% | 1 | 1 | 0% | 1,697 | 1,379 | -19% | 0 | 0 | — |
case-20 | pass→pass | 7,396 | 11,199 | +51% | 1 | 1 | 0% | 1,074 | 2,559 | +138% | 0 | 0 | — |
case-21 | fail→fail | 8,263 | 10,199 | +23% | 1 | 1 | 0% | 341 | 1,765 | +418% | 0 | 0 | — |
case-22 | pass→pass | 13,092 | 7,975 | -39% | 1 | 1 | 0% | 2,042 | 1,943 | -5% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 18 counted toward the lift figure. The other 4 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +50 percentage points is the difference between those two pass rates over the 18 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.