Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Manage opencode permissions: review always-allow lists, suggest safe read-only commands, configure permission patterns
.claude/skills/sickn33-permission-manager/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 11 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -69% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 652% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -26% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -10% | 0% |
Use this when optimizing opencode's permission settings, reviewing allowed commands, or configuring skill access controls.
~/.config/opencode/opencode.json or project-level opencode.jsonComplements opencode's built-in allow/deny/ask permissions by auditing current config and recommending adjustments through conversation.
git status --short, git diff --stat, and ls -lagit status* unless the expanded command family has been manually reviewed as read-onlyUse the Task tool with the permission-manager subagent type:
/permissionsOr in natural language, ask opencode to "manage opencode permissions" or "review allowed commands".
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 7,721 | 2,475 | -68% | 1 | 1 | 0% | 1,446 | 707 | -51% | 0 | 0 | — |
case-02 | fail→pass | 12,011 | 2,594 | -78% | 1 | 1 | 0% | 2,360 | 720 | -69% | 0 | 0 | — |
case-03 | fail→pass | 3,757 | 16,976 | +352% | 1 | 1 | 0% | 487 | 3,663 | +652% | 0 | 0 | — |
case-04 | pass→pass | 11,495 | 10,644 | -7% | 1 | 1 | 0% | 2,167 | 2,289 | +6% | 0 | 0 | — |
case-05 | pass→pass | 7,969 | 5,249 | -34% | 1 | 1 | 0% | 1,429 | 1,408 | -1% | 0 | 0 | — |
case-06 | pass→pass | 11,567 | 9,670 | -16% | 1 | 1 | 0% | 2,112 | 2,278 | +8% | 0 | 0 | — |
case-07 | fail→fail | 6,879 | 4,636 | -33% | 1 | 1 | 0% | 1,280 | 1,249 | -2% | 0 | 0 | — |
case-08 | fail→pass | 10,302 | 4,929 | -52% | 1 | 1 | 0% | 1,689 | 1,258 | -26% | 0 | 0 | — |
case-09 | pass→pass | 13,017 | 7,844 | -40% | 1 | 1 | 0% | 2,177 | 1,654 | -24% | 0 | 0 | — |
case-10 | fail→pass | 9,157 | 4,172 | -54% | 1 | 1 | 0% | 1,627 | 1,099 | -32% | 0 | 0 | — |
case-11 | fail→pass | 9,136 | 6,404 | -30% | 1 | 1 | 0% | 1,661 | 1,492 | -10% | 0 | 0 | — |
case-12 | pass→pass | 10,282 | 4,514 | -56% | 1 | 1 | 0% | 1,890 | 1,226 | -35% | 0 | 0 | — |
case-13 | fail→fail | 8,255 | 3,115 | -62% | 1 | 1 | 0% | 1,303 | 796 | -39% | 0 | 0 | — |
case-14 | pass→pass | 10,154 | 5,113 | -50% | 1 | 1 | 0% | 1,801 | 1,305 | -28% | 0 | 0 | — |
case-15 | fail→pass | 11,548 | 8,118 | -30% | 1 | 1 | 0% | 2,087 | 1,919 | -8% | 0 | 0 | — |
case-16 | pass→pass | 5,018 | 2,525 | -50% | 1 | 1 | 0% | 809 | 797 | -1% | 0 | 0 | — |
case-17 | fail→pass | 13,391 | 5,560 | -58% | 1 | 1 | 0% | 2,154 | 1,264 | -41% | 0 | 0 | — |
case-18 | fail→pass | 9,303 | 3,013 | -68% | 1 | 1 | 0% | 1,520 | 948 | -38% | 0 | 0 | — |
case-19 | fail→pass | 6,201 | 1,461 | -76% | 1 | 1 | 0% | 1,126 | 649 | -42% | 0 | 0 | — |
case-20 | pass→pass | 7,693 | 4,863 | -37% | 1 | 1 | 0% | 1,245 | 1,187 | -5% | 0 | 0 | — |
case-21 | pass→pass | 10,340 | 6,079 | -41% | 1 | 1 | 0% | 1,926 | 1,389 | -28% | 0 | 0 | — |
case-22 | pass→pass | 9,268 | 5,466 | -41% | 1 | 1 | 0% | 1,607 | 1,358 | -15% | 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. The headline lift of +41 percentage points is the difference between those two pass rates over the 22 comparable cases.
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.