Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when adding a new agent harness (CLI-based coding agent) adapter to adapters. Covers capability audit, adapter scaffold, session parsing, auth detection, hooks/plugins wiring, tests, and docs.
.claude/skills/a5c-ai-integrate-harness/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 3063% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -31% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -43% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -12% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -37% | 0% |
Goal: produce a production-quality XAdapter extends BaseAgentAdapter with full test coverage and documentation, matching the level of the existing 11 adapters (claude, codex, cursor, gemini, opencode, openclaw, copilot, hermes, pi, omp, adapters-remote).
AgentCapabilities field. Unknown? Set conservatively (false) and note in PR.packages/adapters/src/<name>-adapter.ts extending BaseAgentAdapter. Required: agent, displayName, cliCommand, minVersion, hostEnvSignals, capabilities, models[], defaultModelId, configSchema, buildSpawnArgs, parseEvent, detectAuth, getAuthGuidance, sessionDir, parseSessionFile, listSessionFiles, readConfig, writeConfig.parseJsonlSessionFile; otherwise write a custom parser and unit-test each event shape.writeNativeHook and mirror into HookConfigManager. If not, rely on the base class's virtual hooks.mcpServers in its config JSON, flip supportsPlugins: true, add pluginFormats: ['mcp-server'], and delegate to mcp-plugins.ts (see cursor/gemini/opencode/openclaw for the pattern).packages/adapters/src/index.ts exports and to the default registry in packages/core/src/client.ts (if applicable).packages/adapters/tests/<name>-adapter.test.ts:buildSpawnArgs for a few representative RunOptionsparseEvent for each JSONL type the harness emitsdetectAuth for authenticated + unauthenticated statesmcp-plugins-parity.test.ts.packages/cli/tests/commands-audit.test.ts passes (it exercises every adapter via the built CLI).local/max-file-lines). Split helpers into sibling modules if you're close.docs/02-agents/<name>.md.npm run changeset, pick minor (new adapter), summarize.bashnpm run typecheck npm run lint npm test npx vitest run packages/adapters/tests/<name>-adapter.test.ts
All existing tests must continue to pass — no regressions in commands-audit.
commands-audit.test.ts won't exercise it.detectAuth() — breaks testability.buildSpawnArgs returning the string "undefined" for missing options — always gate with if (options.X != null).appendJsonHook or appendYamlHook.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 3,507 | 28,926 | +725% | 1 | 1 | 0% | 227 | 7,179 | +3063% | 0 | 0 | — |
case-02 | fail→fail | 4,508 | 4,625 | +3% | 1 | 1 | 0% | 247 | 1,164 | +371% | 0 | 0 | — |
case-03 | fail→fail | 2,794 | 5,408 | +94% | 1 | 1 | 0% | 154 | 987 | +541% | 0 | 0 | — |
case-04 | fail→fail | 10,689 | 4,959 | -54% | 1 | 1 | 0% | 2,192 | 1,001 | -54% | 0 | 0 | — |
case-05 | fail→fail | 11,910 | 5,028 | -58% | 1 | 1 | 0% | 1,934 | 1,081 | -44% | 0 | 0 | — |
case-06 | fail→fail | 14,134 | 4,057 | -71% | 1 | 1 | 0% | 2,829 | 1,000 | -65% | 0 | 0 | — |
case-07 | fail→pass | 9,671 | 1,716 | -82% | 1 | 1 | 0% | 1,534 | 1,052 | -31% | 0 | 0 | — |
case-08 | fail→fail | 11,793 | 5,507 | -53% | 1 | 1 | 0% | 2,245 | 1,840 | -18% | 0 | 0 | — |
case-09 | fail→fail | 12,848 | 5,137 | -60% | 1 | 1 | 0% | 2,474 | 1,758 | -29% | 0 | 0 | — |
case-10 | fail→pass | 15,491 | 4,712 | -70% | 1 | 1 | 0% | 2,968 | 1,689 | -43% | 0 | 0 | — |
case-11 | fail→pass | 7,584 | 2,543 | -66% | 1 | 1 | 0% | 1,385 | 1,222 | -12% | 0 | 0 | — |
case-12 | fail→pass | 17,750 | 5,145 | -71% | 1 | 1 | 0% | 2,750 | 1,725 | -37% | 0 | 0 | — |
case-13 | fail→pass | 15,933 | 6,885 | -57% | 1 | 1 | 0% | 2,469 | 1,807 | -27% | 0 | 0 | — |
case-14 | fail→pass | 4,489 | 2,359 | -47% | 1 | 1 | 0% | 763 | 1,261 | +65% | 0 | 0 | — |
case-15 | fail→fail | 5,367 | 1,529 | -72% | 1 | 1 | 0% | 880 | 1,026 | +17% | 0 | 0 | — |
case-16 | fail→pass | 9,715 | 2,874 | -70% | 1 | 1 | 0% | 1,725 | 1,293 | -25% | 0 | 0 | — |
case-17 | fail→pass | 8,693 | 3,871 | -55% | 1 | 1 | 0% | 1,581 | 1,620 | +2% | 0 | 0 | — |
case-18 | fail→pass | 13,886 | 7,362 | -47% | 1 | 1 | 0% | 2,188 | 2,068 | -5% | 0 | 0 | — |
case-19 | fail→fail | 14,258 | 1,450 | -90% | 1 | 1 | 0% | 2,576 | 1,034 | -60% | 0 | 0 | — |
case-20 | fail→fail | 11,919 | 2,525 | -79% | 1 | 1 | 0% | 1,714 | 1,211 | -29% | 0 | 0 | — |
case-21 | fail→fail | 10,500 | 1,710 | -84% | 1 | 1 | 0% | 1,723 | 1,084 | -37% | 0 | 0 | — |
case-22 | fail→fail | 9,588 | 1,177 | -88% | 1 | 1 | 0% | 1,733 | 962 | -44% | 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 16 counted toward the lift figure. The other 6 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 +45 percentage points is the difference between those two pass rates over the 16 comparable cases. 3 cases got worse with the skill loaded, and they are 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.