@vercel/agent-eval 2.4.1 → 2.6.0

package/dist/lib/agents/gemini/agent.js.map

@@ -1 +1 @@
-{"version":3,"file":"agent.js","sourceRoot":"","sources":["../../../../src/lib/agents/gemini/agent.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AAIzC,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAE7C,OAAO,EAAE,iBAAiB,EAAE,MAAM,2BAA2B,CAAC;AAE9D;;;;;;GAMG;AACH,MAAM,UAAU,sBAAsB;IACpC,OAAO;QACL,IAAI,EAAE,QAAQ;QACd,WAAW,EAAE,YAAY;QACzB,YAAY,EAAE,sBAAsB;QACpC,aAAa,EAAE,QAAQ;QACvB,+EAA+E;QAC/E,6DAA6D;QAC7D,UAAU,EAAE,aAAa,CAAC,IAAI,GAAG,CAAC,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAEhE,eAAe;YACb,OAAO,aAAa,CAAC,YAAY,CAAC;QACpC,CAAC;QAED,OAAO,CAAC,QAAyB;YAC/B,6DAA6D;YAC7D,kDAAkD;YAClD,yEAAyE;YACzE,0DAA0D;YAC1D,OAAO;gBACL,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,SAAS,CAAC,EAAE,SAAS,EAAE,IAAI,EAAE,WAAW,EAAE,oBAAoB,EAAE,SAAS,EAAE,QAAQ,EAAE;gBAC3H,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,SAAS,EAAE,IAAI,EAAE,oBAAoB,CAAC,EAAE,WAAW,EAAE,2BAA2B,EAAE,SAAS,EAAE,QAAQ,EAAE;aAC9I,CAAC;QACJ,CAAC;QAED,6DAA6D;QAC7D,WAAW;YACT,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,OAAO,CAAC,OAAwB;YAC9B,OAAO,EAAE,CAAC,aAAa,CAAC,YAAY,CAAC,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC;QAC1D,CAAC;KACF,CAAC;AACJ,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,iBAAiB;IAC/B,MAAM,UAAU,GAAG,sBAAsB,EAAE,CAAC;IAC5C,OAAO;QACL,IAAI,EAAE,UAAU,CAAC,IAAI;QACrB,WAAW,EAAE,UAAU,CAAC,WAAW;QACnC,eAAe,EAAE,UAAU,CAAC,eAAe;QAC3C,eAAe;YACb,OAAO,UAAU,CAAC,YAAY,CAAC;QACjC,CAAC;QACD,GAAG,EAAE,CAAC,WAAW,EAAE,OAAO,EAAE,EAAE,CAAC,iBAAiB,CAAC,UAAU,EAAE,WAAW,EAAE,OAAO,CAAC;QAClF,UAAU;KACX,CAAC;AACJ,CAAC"}
+{"version":3,"file":"agent.js","sourceRoot":"","sources":["../../../../src/lib/agents/gemini/agent.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AAIzC,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAE7C,OAAO,EAAE,iBAAiB,EAAE,MAAM,2BAA2B,CAAC;AAE9D;;;;;;GAMG;AACH,MAAM,UAAU,sBAAsB;IACpC,OAAO;QACL,IAAI,EAAE,QAAQ;QACd,WAAW,EAAE,YAAY;QACzB,YAAY,EAAE,sBAAsB;QACpC,aAAa,EAAE,QAAQ;QACvB,+EAA+E;QAC/E,6DAA6D;QAC7D,UAAU,EAAE,aAAa,CAAC,IAAI,GAAG,CAAC,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAEhE,eAAe;YACb,OAAO,aAAa,CAAC,YAAY,CAAC;QACpC,CAAC;QAED,OAAO,CAAC,QAAyB;YAC/B,6DAA6D;YAC7D,kDAAkD;YAClD,yEAAyE;YACzE,0DAA0D;YAC1D,OAAO;gBACL,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,SAAS,CAAC,EAAE,SAAS,EAAE,IAAI,EAAE,WAAW,EAAE,oBAAoB,EAAE,SAAS,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE;gBAC7I,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,SAAS,EAAE,IAAI,EAAE,oBAAoB,CAAC,EAAE,WAAW,EAAE,2BAA2B,EAAE,SAAS,EAAE,QAAQ,EAAE;aAC9I,CAAC;QACJ,CAAC;QAED,6DAA6D;QAC7D,WAAW;YACT,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,0CAA0C;QAC1C,cAAc;YACZ,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,WAAW,CAAC,EAAE,CAAC;QACjE,CAAC;QAED,OAAO,CAAC,OAAwB;YAC9B,OAAO,EAAE,CAAC,aAAa,CAAC,YAAY,CAAC,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC;QAC1D,CAAC;KACF,CAAC;AACJ,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,iBAAiB;IAC/B,MAAM,UAAU,GAAG,sBAAsB,EAAE,CAAC;IAC5C,OAAO;QACL,IAAI,EAAE,UAAU,CAAC,IAAI;QACrB,WAAW,EAAE,UAAU,CAAC,WAAW;QACnC,eAAe,EAAE,UAAU,CAAC,eAAe;QAC3C,eAAe;YACb,OAAO,UAAU,CAAC,YAAY,CAAC;QACjC,CAAC;QACD,GAAG,EAAE,CAAC,WAAW,EAAE,OAAO,EAAE,EAAE,CAAC,iBAAiB,CAAC,UAAU,EAAE,WAAW,EAAE,OAAO,CAAC;QAClF,UAAU;KACX,CAAC;AACJ,CAAC"}

package/dist/lib/agents/index.d.ts

@@ -9,5 +9,6 @@ export declare function assertRunRuntimeControls(agentName: AgentType, controls:
     webResearch?: boolean;
 }, judgeAgentName?: AgentType): void;
 export { registerAgent, getAgent, listAgents, hasAgent };
+export { runCredentials } from './plugin/orchestrator.js';
 export type { Agent, AgentRunOptions, AgentRunResult } from './types.js';
 //# sourceMappingURL=index.d.ts.map

package/dist/lib/agents/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/lib/agents/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,aAAa,EAAE,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,eAAe,CAAC;AAY9E,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAY7C,sEAAsE;AACtE,wBAAgB,wBAAwB,CACtC,SAAS,EAAE,SAAS,EACpB,QAAQ,EAAE;IAAE,oBAAoB,CAAC,EAAE,OAAO,CAAC;IAAC,WAAW,CAAC,EAAE,OAAO,CAAA;CAAE,EACnE,cAAc,CAAC,EAAE,SAAS,GACzB,IAAI,CAUN;AAGD,OAAO,EAAE,aAAa,EAAE,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC;AAGzD,YAAY,EAAE,KAAK,EAAE,eAAe,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/lib/agents/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,aAAa,EAAE,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,eAAe,CAAC;AAa9E,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAa7C,sEAAsE;AACtE,wBAAgB,wBAAwB,CACtC,SAAS,EAAE,SAAS,EACpB,QAAQ,EAAE;IAAE,oBAAoB,CAAC,EAAE,OAAO,CAAC;IAAC,WAAW,CAAC,EAAE,OAAO,CAAA;CAAE,EACnE,cAAc,CAAC,EAAE,SAAS,GACzB,IAAI,CAUN;AAGD,OAAO,EAAE,aAAa,EAAE,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC;AAGzD,OAAO,EAAE,cAAc,EAAE,MAAM,0BAA0B,CAAC;AAG1D,YAAY,EAAE,KAAK,EAAE,eAAe,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC"}

package/dist/lib/agents/index.js

@@ -8,6 +8,7 @@ import { createOpenCodeAgent } from './opencode/agent.js';
 import { createFxAgent } from './fx/agent.js';
 import { createGeminiAgent } from './gemini/agent.js';
 import { createCursorAgent } from './cursor/agent.js';
+import { createPiCodingAgent } from './pi-coding-agent/agent.js';
 import { assertBundledSkillsControl, assertCrossAgentJudgeSupport, assertWebResearchControl, } from './plugin/contract.js';
 // Register all agent variants (Vercel AI Gateway + Direct API)
 registerAgent(createClaudeCodeAgent({ useVercelAiGateway: true })); // vercel-ai-gateway/claude-code
@@ -18,6 +19,7 @@ registerAgent(createOpenCodeAgent()); // vercel-ai-gateway/opencode
 registerAgent(createFxAgent()); // vercel-ai-gateway/fx
 registerAgent(createGeminiAgent()); // gemini
 registerAgent(createCursorAgent()); // cursor
+registerAgent(createPiCodingAgent()); // vercel-ai-gateway/pi-coding-agent
 /** Validate opt-in runtime controls for every agent used by a run. */
 export function assertRunRuntimeControls(agentName, controls, judgeAgentName) {
     const definition = getAgent(agentName).definition;
@@ -31,4 +33,6 @@ export function assertRunRuntimeControls(agentName, controls, judgeAgentName) {
 }
 // Re-export registry functions
 export { registerAgent, getAgent, listAgents, hasAgent };
+// Credentials a run injects, for redacting payloads built outside the agent.
+export { runCredentials } from './plugin/orchestrator.js';
 //# sourceMappingURL=index.js.map

package/dist/lib/agents/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/lib/agents/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,aAAa,EAAE,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,eAAe,CAAC;AAC9E,OAAO,EAAE,qBAAqB,EAAE,MAAM,wBAAwB,CAAC;AAC/D,OAAO,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AACpD,OAAO,EAAE,mBAAmB,EAAE,MAAM,qBAAqB,CAAC;AAC1D,OAAO,EAAE,aAAa,EAAE,MAAM,eAAe,CAAC;AAC9C,OAAO,EAAE,iBAAiB,EAAE,MAAM,mBAAmB,CAAC;AACtD,OAAO,EAAE,iBAAiB,EAAE,MAAM,mBAAmB,CAAC;AACtD,OAAO,EACL,0BAA0B,EAC1B,4BAA4B,EAC5B,wBAAwB,GACzB,MAAM,sBAAsB,CAAC;AAG9B,+DAA+D;AAC/D,aAAa,CAAC,qBAAqB,CAAC,EAAE,kBAAkB,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAG,gCAAgC;AACtG,aAAa,CAAC,qBAAqB,CAAC,EAAE,kBAAkB,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,CAAE,cAAc;AACpF,aAAa,CAAC,gBAAgB,CAAC,EAAE,kBAAkB,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAQ,0BAA0B;AAChG,aAAa,CAAC,gBAAgB,CAAC,EAAE,kBAAkB,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,CAAO,QAAQ;AAC9E,aAAa,CAAC,mBAAmB,EAAE,CAAC,CAAC,CAAiC,6BAA6B;AACnG,aAAa,CAAC,aAAa,EAAE,CAAC,CAAC,CAAuC,uBAAuB;AAC7F,aAAa,CAAC,iBAAiB,EAAE,CAAC,CAAC,CAAmC,SAAS;AAC/E,aAAa,CAAC,iBAAiB,EAAE,CAAC,CAAC,CAAmC,SAAS;AAE/E,sEAAsE;AACtE,MAAM,UAAU,wBAAwB,CACtC,SAAoB,EACpB,QAAmE,EACnE,cAA0B;IAE1B,MAAM,UAAU,GAAG,QAAQ,CAAC,SAAS,CAAC,CAAC,UAAU,CAAC;IAClD,0BAA0B,CAAC,UAAU,EAAE,QAAQ,CAAC,oBAAoB,CAAC,CAAC;IACtE,wBAAwB,CAAC,UAAU,EAAE,QAAQ,CAAC,WAAW,CAAC,CAAC;IAE3D,IAAI,cAAc,IAAI,cAAc,KAAK,SAAS,EAAE,CAAC;QACnD,MAAM,eAAe,GAAG,QAAQ,CAAC,cAAc,CAAC,CAAC,UAAU,CAAC;QAC5D,0BAA0B,CAAC,eAAe,EAAE,QAAQ,CAAC,oBAAoB,CAAC,CAAC;QAC3E,4BAA4B,CAAC,eAAe,CAAC,CAAC;IAChD,CAAC;AACH,CAAC;AAED,+BAA+B;AAC/B,OAAO,EAAE,aAAa,EAAE,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/lib/agents/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,aAAa,EAAE,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,eAAe,CAAC;AAC9E,OAAO,EAAE,qBAAqB,EAAE,MAAM,wBAAwB,CAAC;AAC/D,OAAO,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AACpD,OAAO,EAAE,mBAAmB,EAAE,MAAM,qBAAqB,CAAC;AAC1D,OAAO,EAAE,aAAa,EAAE,MAAM,eAAe,CAAC;AAC9C,OAAO,EAAE,iBAAiB,EAAE,MAAM,mBAAmB,CAAC;AACtD,OAAO,EAAE,iBAAiB,EAAE,MAAM,mBAAmB,CAAC;AACtD,OAAO,EAAE,mBAAmB,EAAE,MAAM,4BAA4B,CAAC;AACjE,OAAO,EACL,0BAA0B,EAC1B,4BAA4B,EAC5B,wBAAwB,GACzB,MAAM,sBAAsB,CAAC;AAG9B,+DAA+D;AAC/D,aAAa,CAAC,qBAAqB,CAAC,EAAE,kBAAkB,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAG,gCAAgC;AACtG,aAAa,CAAC,qBAAqB,CAAC,EAAE,kBAAkB,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,CAAE,cAAc;AACpF,aAAa,CAAC,gBAAgB,CAAC,EAAE,kBAAkB,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAQ,0BAA0B;AAChG,aAAa,CAAC,gBAAgB,CAAC,EAAE,kBAAkB,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,CAAO,QAAQ;AAC9E,aAAa,CAAC,mBAAmB,EAAE,CAAC,CAAC,CAAiC,6BAA6B;AACnG,aAAa,CAAC,aAAa,EAAE,CAAC,CAAC,CAAuC,uBAAuB;AAC7F,aAAa,CAAC,iBAAiB,EAAE,CAAC,CAAC,CAAmC,SAAS;AAC/E,aAAa,CAAC,iBAAiB,EAAE,CAAC,CAAC,CAAmC,SAAS;AAC/E,aAAa,CAAC,mBAAmB,EAAE,CAAC,CAAC,CAAiC,oCAAoC;AAE1G,sEAAsE;AACtE,MAAM,UAAU,wBAAwB,CACtC,SAAoB,EACpB,QAAmE,EACnE,cAA0B;IAE1B,MAAM,UAAU,GAAG,QAAQ,CAAC,SAAS,CAAC,CAAC,UAAU,CAAC;IAClD,0BAA0B,CAAC,UAAU,EAAE,QAAQ,CAAC,oBAAoB,CAAC,CAAC;IACtE,wBAAwB,CAAC,UAAU,EAAE,QAAQ,CAAC,WAAW,CAAC,CAAC;IAE3D,IAAI,cAAc,IAAI,cAAc,KAAK,SAAS,EAAE,CAAC;QACnD,MAAM,eAAe,GAAG,QAAQ,CAAC,cAAc,CAAC,CAAC,UAAU,CAAC;QAC5D,0BAA0B,CAAC,eAAe,EAAE,QAAQ,CAAC,oBAAoB,CAAC,CAAC;QAC3E,4BAA4B,CAAC,eAAe,CAAC,CAAC;IAChD,CAAC;AACH,CAAC;AAED,+BAA+B;AAC/B,OAAO,EAAE,aAAa,EAAE,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC;AAEzD,6EAA6E;AAC7E,OAAO,EAAE,cAAc,EAAE,MAAM,0BAA0B,CAAC"}

package/dist/lib/agents/opencode/agent.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"agent.d.ts","sourceRoot":"","sources":["../../../../src/lib/agents/opencode/agent.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAIH,OAAO,KAAK,EAAE,KAAK,EAAmB,MAAM,aAAa,CAAC;AAG1D,OAAO,KAAK,EAAE,eAAe,EAA2B,MAAM,uBAAuB,CAAC;AAGtF;;;;;GAKG;AACH,MAAM,WAAW,sBAAsB;IACrC,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAClC,MAAM,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE;QACtB,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,SAAS,CAAC,EAAE,OAAO,CAAC;QACpB,SAAS,CAAC,EAAE,OAAO,CAAC;QACpB,UAAU,CAAC,EAAE,OAAO,CAAC;QACrB,WAAW,CAAC,EAAE,OAAO,CAAC;QACtB,KAAK,CAAC,EAAE;YAAE,OAAO,EAAE,MAAM,CAAC;YAAC,MAAM,EAAE,MAAM,CAAA;SAAE,CAAC;KAC7C,CAAC,CAAC;CACJ;AAED;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,wBAAgB,oBAAoB,CAClC,KAAK,EAAE,MAAM,EACb,cAAc,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,sBAAsB,CAAC,GACtD,MAAM,CAMR;AAED;;;;;;;;GAQG;AACH,wBAAgB,sBAAsB,CACpC,cAAc,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,sBAAsB,CAAC,EACvD,MAAM,CAAC,EAAE,MAAM,EACf,SAAS,CAAC,EAAE,MAAM,GACjB,MAAM,CA0BR;AAED;;;;;;;;GAQG;AACH,wBAAgB,wBAAwB,IAAI,eAAe,CA2G1D;AAED;;;GAGG;AACH,wBAAgB,mBAAmB,IAAI,KAAK,CAY3C"}
+{"version":3,"file":"agent.d.ts","sourceRoot":"","sources":["../../../../src/lib/agents/opencode/agent.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAIH,OAAO,KAAK,EAAE,KAAK,EAAmB,MAAM,aAAa,CAAC;AAG1D,OAAO,KAAK,EAAE,eAAe,EAA2B,MAAM,uBAAuB,CAAC;AAGtF;;;;;GAKG;AACH,MAAM,WAAW,sBAAsB;IACrC,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAClC,MAAM,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE;QACtB,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,SAAS,CAAC,EAAE,OAAO,CAAC;QACpB,SAAS,CAAC,EAAE,OAAO,CAAC;QACpB,UAAU,CAAC,EAAE,OAAO,CAAC;QACrB,WAAW,CAAC,EAAE,OAAO,CAAC;QACtB,KAAK,CAAC,EAAE;YAAE,OAAO,EAAE,MAAM,CAAC;YAAC,MAAM,EAAE,MAAM,CAAA;SAAE,CAAC;KAC7C,CAAC,CAAC;CACJ;AAED;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,wBAAgB,oBAAoB,CAClC,KAAK,EAAE,MAAM,EACb,cAAc,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,sBAAsB,CAAC,GACtD,MAAM,CAMR;AAED;;;;;;;;GAQG;AACH,wBAAgB,sBAAsB,CACpC,cAAc,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,sBAAsB,CAAC,EACvD,MAAM,CAAC,EAAE,MAAM,EACf,SAAS,CAAC,EAAE,MAAM,GACjB,MAAM,CA0BR;AAED;;;;;;;;GAQG;AACH,wBAAgB,wBAAwB,IAAI,eAAe,CAkH1D;AAED;;;GAGG;AACH,wBAAgB,mBAAmB,IAAI,KAAK,CAY3C"}

package/dist/lib/agents/opencode/agent.js

@@ -121,6 +121,7 @@ export function createOpenCodeDefinition() {
                     retryOnce: true,
                     errorPrefix: 'npm install failed',
                     errorBody: 'last10',
+                    scope: 'project',
                 },
             ];
             if (binaryUrl) {
@@ -153,6 +154,11 @@ export function createOpenCodeDefinition() {
             }
             return steps;
         },
+        // The same `opencode` on PATH that run.mjs invokes (npm or a custom binaryUrl
+        // build); recorded as provenance.agentCliVersion.
+        versionCommand() {
+            return { kind: 'command', cmd: 'opencode', args: ['--version'] };
+        },
         // OpenCode is configured via a project-local opencode.json (writeFiles target,
         // relative to cwd — not a `~` path, so no viaShell heredoc needed).
         configFiles(options) {

package/dist/lib/agents/opencode/agent.js.map

@@ -1 +1 @@
-{"version":3,"file":"agent.js","sourceRoot":"","sources":["../../../../src/lib/agents/opencode/agent.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AAIzC,OAAO,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAE1C,OAAO,EAAE,iBAAiB,EAAE,MAAM,2BAA2B,CAAC;AAqB9D;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,MAAM,UAAU,oBAAoB,CAClC,KAAa,EACb,cAAuD;IAEvD,MAAM,UAAU,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;IACvC,IAAI,UAAU,KAAK,QAAQ,IAAI,CAAC,cAAc,IAAI,MAAM,CAAC,MAAM,CAAC,cAAc,EAAE,UAAU,CAAC,CAAC,EAAE,CAAC;QAC7F,OAAO,KAAK,CAAC;IACf,CAAC;IACD,OAAO,UAAU,KAAK,EAAE,CAAC;AAC3B,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,sBAAsB,CACpC,cAAuD,EACvD,MAAe,EACf,SAAkB;IAElB,MAAM,UAAU,GAA4B;QAC1C,OAAO,EAAE;YACP,MAAM,EAAE,MAAM,IAAI,0BAA0B;YAC5C,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAC7C;KACF,CAAC;IACF,MAAM,EAAE,MAAM,EAAE,WAAW,EAAE,GAAG,cAAc,EAAE,GAAG,cAAc,IAAI,EAAE,CAAC;IAExE,MAAM,SAAS,GAA4B;QACzC,MAAM,EAAE;YACN,GAAG,UAAU;YACb,GAAG,WAAW;YACd,OAAO,EAAE,EAAE,GAAI,UAAU,CAAC,OAAmC,EAAE,GAAG,WAAW,EAAE,OAAO,EAAE;SACzF;QACD,GAAG,cAAc;KAClB,CAAC;IAEF,OAAO,IAAI,CAAC,SAAS,CAAC;QACpB,OAAO,EAAE,iCAAiC;QAC1C,QAAQ,EAAE,SAAS;QACnB,sEAAsE;QACtE,uEAAuE;QACvE,+DAA+D;QAC/D,UAAU,EAAE,OAAO;KACpB,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;AACd,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,wBAAwB;IACtC,OAAO;QACL,IAAI,EAAE,4BAA4B;QAClC,WAAW,EAAE,8BAA8B;QAC3C,YAAY,EAAE,kCAAkC;QAChD,aAAa,EAAE,4BAA4B;QAC3C,oBAAoB,EAAE,gBAAgB;QACtC,+EAA+E;QAC/E,6DAA6D;QAC7D,UAAU,EAAE,aAAa,CAAC,IAAI,GAAG,CAAC,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAEhE,4EAA4E;QAC5E,eAAe;YACb,OAAO,UAAU,CAAC,YAAY,CAAC;QACjC,CAAC;QAED,OAAO,CAAC,OAAwB;YAC9B,6EAA6E;YAC7E,0EAA0E;YAC1E,WAAW;YACX,yEAAyE;YACzE,yEAAyE;YACzE,0EAA0E;YAC1E,MAAM,SAAS,GAAG,OAAO,CAAC,YAAY,EAAE,SAA+B,CAAC;YAExE,MAAM,KAAK,GAAkB;gBAC3B;oBACE,IAAI,EAAE,SAAS;oBACf,GAAG,EAAE,KAAK;oBACV,IAAI,EAAE,CAAC,SAAS,CAAC;oBACjB,SAAS,EAAE,IAAI;oBACf,WAAW,EAAE,oBAAoB;oBACjC,SAAS,EAAE,QAAQ;iBACpB;aACF,CAAC;YAEF,IAAI,SAAS,EAAE,CAAC;gBACd,yEAAyE;gBACzE,2EAA2E;gBAC3E,0EAA0E;gBAC1E,oEAAoE;gBACpE,sEAAsE;gBACtE,0EAA0E;gBAC1E,6EAA6E;gBAC7E,KAAK,CAAC,IAAI,CAAC;oBACT,IAAI,EAAE,SAAS;oBACf,GAAG,EAAE,MAAM;oBACX,IAAI,EAAE;wBACJ,IAAI;wBACJ,4CAA4C,SAAS,sEAAsE;qBAC5H;oBACD,WAAW,EAAE,6BAA6B;oBAC1C,SAAS,EAAE,QAAQ;iBACpB,CAAC,CAAC;YACL,CAAC;iBAAM,CAAC;gBACN,KAAK,CAAC,IAAI,CAAC;oBACT,IAAI,EAAE,SAAS;oBACf,GAAG,EAAE,KAAK;oBACV,IAAI,EAAE,CAAC,SAAS,EAAE,IAAI,EAAE,aAAa,CAAC;oBACtC,WAAW,EAAE,6BAA6B;oBAC1C,SAAS,EAAE,QAAQ;iBACpB,CAAC,CAAC;YACL,CAAC;YAED,OAAO,KAAK,CAAC;QACf,CAAC;QAED,+EAA+E;QAC/E,oEAAoE;QACpE,WAAW,CAAC,OAAwB;YAClC,MAAM,cAAc,GAAG,OAAO,CAAC,YAAY,EAAE,cAEhC,CAAC;YACd,MAAM,OAAO,GAAG,sBAAsB,CACpC,cAAc,EACd,OAAO,CAAC,MAAM,EACd,OAAO,CAAC,OAAO,CAChB,CAAC;YACF,OAAO,CAAC,EAAE,IAAI,EAAE,eAAe,EAAE,OAAO,EAAE,CAAC,CAAC;QAC9C,CAAC;QAED,OAAO,CAAC,OAAwB;YAC9B,OAAO;gBACL,CAAC,UAAU,CAAC,YAAY,CAAC,EAAE,OAAO,CAAC,MAAM;gBACzC,yEAAyE;gBACzE,kEAAkE;gBAClE,0BAA0B;gBAC1B,GAAG,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,mBAAmB,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC7D,CAAC;QACJ,CAAC;QAED;;;;;;WAMG;QACH,WAAW,CAAC,OAAwB;YAClC,MAAM,cAAc,GAAG,OAAO,CAAC,YAAY,EAAE,cAEhC,CAAC;YACd,OAAO;gBACL,QAAQ,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,oBAAoB,CAAC,OAAO,CAAC,KAAK,EAAE,cAAc,CAAC,CAAC,CAAC,CAAC,IAAI;aACrF,CAAC;QACJ,CAAC;KACF,CAAC;AACJ,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,mBAAmB;IACjC,MAAM,UAAU,GAAG,wBAAwB,EAAE,CAAC;IAC9C,OAAO;QACL,IAAI,EAAE,UAAU,CAAC,IAAI;QACrB,WAAW,EAAE,UAAU,CAAC,WAAW;QACnC,eAAe,EAAE,UAAU,CAAC,eAAe;QAC3C,eAAe;YACb,OAAO,UAAU,CAAC,YAAY,CAAC;QACjC,CAAC;QACD,GAAG,EAAE,CAAC,WAAW,EAAE,OAAO,EAAE,EAAE,CAAC,iBAAiB,CAAC,UAAU,EAAE,WAAW,EAAE,OAAO,CAAC;QAClF,UAAU;KACX,CAAC;AACJ,CAAC"}
+{"version":3,"file":"agent.js","sourceRoot":"","sources":["../../../../src/lib/agents/opencode/agent.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AAIzC,OAAO,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAE1C,OAAO,EAAE,iBAAiB,EAAE,MAAM,2BAA2B,CAAC;AAqB9D;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,MAAM,UAAU,oBAAoB,CAClC,KAAa,EACb,cAAuD;IAEvD,MAAM,UAAU,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;IACvC,IAAI,UAAU,KAAK,QAAQ,IAAI,CAAC,cAAc,IAAI,MAAM,CAAC,MAAM,CAAC,cAAc,EAAE,UAAU,CAAC,CAAC,EAAE,CAAC;QAC7F,OAAO,KAAK,CAAC;IACf,CAAC;IACD,OAAO,UAAU,KAAK,EAAE,CAAC;AAC3B,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,sBAAsB,CACpC,cAAuD,EACvD,MAAe,EACf,SAAkB;IAElB,MAAM,UAAU,GAA4B;QAC1C,OAAO,EAAE;YACP,MAAM,EAAE,MAAM,IAAI,0BAA0B;YAC5C,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAC7C;KACF,CAAC;IACF,MAAM,EAAE,MAAM,EAAE,WAAW,EAAE,GAAG,cAAc,EAAE,GAAG,cAAc,IAAI,EAAE,CAAC;IAExE,MAAM,SAAS,GAA4B;QACzC,MAAM,EAAE;YACN,GAAG,UAAU;YACb,GAAG,WAAW;YACd,OAAO,EAAE,EAAE,GAAI,UAAU,CAAC,OAAmC,EAAE,GAAG,WAAW,EAAE,OAAO,EAAE;SACzF;QACD,GAAG,cAAc;KAClB,CAAC;IAEF,OAAO,IAAI,CAAC,SAAS,CAAC;QACpB,OAAO,EAAE,iCAAiC;QAC1C,QAAQ,EAAE,SAAS;QACnB,sEAAsE;QACtE,uEAAuE;QACvE,+DAA+D;QAC/D,UAAU,EAAE,OAAO;KACpB,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;AACd,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,wBAAwB;IACtC,OAAO;QACL,IAAI,EAAE,4BAA4B;QAClC,WAAW,EAAE,8BAA8B;QAC3C,YAAY,EAAE,kCAAkC;QAChD,aAAa,EAAE,4BAA4B;QAC3C,oBAAoB,EAAE,gBAAgB;QACtC,+EAA+E;QAC/E,6DAA6D;QAC7D,UAAU,EAAE,aAAa,CAAC,IAAI,GAAG,CAAC,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAEhE,4EAA4E;QAC5E,eAAe;YACb,OAAO,UAAU,CAAC,YAAY,CAAC;QACjC,CAAC;QAED,OAAO,CAAC,OAAwB;YAC9B,6EAA6E;YAC7E,0EAA0E;YAC1E,WAAW;YACX,yEAAyE;YACzE,yEAAyE;YACzE,0EAA0E;YAC1E,MAAM,SAAS,GAAG,OAAO,CAAC,YAAY,EAAE,SAA+B,CAAC;YAExE,MAAM,KAAK,GAAkB;gBAC3B;oBACE,IAAI,EAAE,SAAS;oBACf,GAAG,EAAE,KAAK;oBACV,IAAI,EAAE,CAAC,SAAS,CAAC;oBACjB,SAAS,EAAE,IAAI;oBACf,WAAW,EAAE,oBAAoB;oBACjC,SAAS,EAAE,QAAQ;oBACnB,KAAK,EAAE,SAAS;iBACjB;aACF,CAAC;YAEF,IAAI,SAAS,EAAE,CAAC;gBACd,yEAAyE;gBACzE,2EAA2E;gBAC3E,0EAA0E;gBAC1E,oEAAoE;gBACpE,sEAAsE;gBACtE,0EAA0E;gBAC1E,6EAA6E;gBAC7E,KAAK,CAAC,IAAI,CAAC;oBACT,IAAI,EAAE,SAAS;oBACf,GAAG,EAAE,MAAM;oBACX,IAAI,EAAE;wBACJ,IAAI;wBACJ,4CAA4C,SAAS,sEAAsE;qBAC5H;oBACD,WAAW,EAAE,6BAA6B;oBAC1C,SAAS,EAAE,QAAQ;iBACpB,CAAC,CAAC;YACL,CAAC;iBAAM,CAAC;gBACN,KAAK,CAAC,IAAI,CAAC;oBACT,IAAI,EAAE,SAAS;oBACf,GAAG,EAAE,KAAK;oBACV,IAAI,EAAE,CAAC,SAAS,EAAE,IAAI,EAAE,aAAa,CAAC;oBACtC,WAAW,EAAE,6BAA6B;oBAC1C,SAAS,EAAE,QAAQ;iBACpB,CAAC,CAAC;YACL,CAAC;YAED,OAAO,KAAK,CAAC;QACf,CAAC;QAED,8EAA8E;QAC9E,kDAAkD;QAClD,cAAc;YACZ,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,UAAU,EAAE,IAAI,EAAE,CAAC,WAAW,CAAC,EAAE,CAAC;QACnE,CAAC;QAED,+EAA+E;QAC/E,oEAAoE;QACpE,WAAW,CAAC,OAAwB;YAClC,MAAM,cAAc,GAAG,OAAO,CAAC,YAAY,EAAE,cAEhC,CAAC;YACd,MAAM,OAAO,GAAG,sBAAsB,CACpC,cAAc,EACd,OAAO,CAAC,MAAM,EACd,OAAO,CAAC,OAAO,CAChB,CAAC;YACF,OAAO,CAAC,EAAE,IAAI,EAAE,eAAe,EAAE,OAAO,EAAE,CAAC,CAAC;QAC9C,CAAC;QAED,OAAO,CAAC,OAAwB;YAC9B,OAAO;gBACL,CAAC,UAAU,CAAC,YAAY,CAAC,EAAE,OAAO,CAAC,MAAM;gBACzC,yEAAyE;gBACzE,kEAAkE;gBAClE,0BAA0B;gBAC1B,GAAG,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,mBAAmB,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC7D,CAAC;QACJ,CAAC;QAED;;;;;;WAMG;QACH,WAAW,CAAC,OAAwB;YAClC,MAAM,cAAc,GAAG,OAAO,CAAC,YAAY,EAAE,cAEhC,CAAC;YACd,OAAO;gBACL,QAAQ,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,oBAAoB,CAAC,OAAO,CAAC,KAAK,EAAE,cAAc,CAAC,CAAC,CAAC,CAAC,IAAI;aACrF,CAAC;QACJ,CAAC;KACF,CAAC;AACJ,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,mBAAmB;IACjC,MAAM,UAAU,GAAG,wBAAwB,EAAE,CAAC;IAC9C,OAAO;QACL,IAAI,EAAE,UAAU,CAAC,IAAI;QACrB,WAAW,EAAE,UAAU,CAAC,WAAW;QACnC,eAAe,EAAE,UAAU,CAAC,eAAe;QAC3C,eAAe;YACb,OAAO,UAAU,CAAC,YAAY,CAAC;QACjC,CAAC;QACD,GAAG,EAAE,CAAC,WAAW,EAAE,OAAO,EAAE,EAAE,CAAC,iBAAiB,CAAC,UAAU,EAAE,WAAW,EAAE,OAAO,CAAC;QAClF,UAAU;KACX,CAAC;AACJ,CAAC"}

package/dist/lib/agents/pi-coding-agent/agent.d.ts

@@ -0,0 +1,61 @@
+/**
+ * PI coding agent (`pi`, @earendil-works/pi-coding-agent) — host-side definition +
+ * the thin Agent wrapper.
+ *
+ * The definition is pure data/auth/config; the actual CLI invocation + transcript
+ * capture live in ./run.mjs (shipped into the sandbox by the orchestrator). PI
+ * only ever runs as a child process of that runner, inside the sandbox — its SDK
+ * is never imported into the evaluator process, so a PI failure is an agent-run
+ * failure, not an evaluator crash.
+ *
+ * PI is multi-provider, but like OpenCode it only runs through the Vercel AI
+ * Gateway here (PI's built-in `vercel-ai-gateway` provider), so there is no
+ * gateway/direct fork: a single agent, always authenticated with
+ * AI_GATEWAY_API_KEY. See {@link PiAgentOptions} for what agentOptions accepts.
+ */
+import type { Agent } from '../types.js';
+import type { AgentDefinition } from '../plugin/contract.js';
+/**
+ * Additional provider configuration for PI's `models.json`, e.g. gateway models
+ * PI's built-in catalog does not list yet. Passed through
+ * agentOptions.extraProviders verbatim — see PI's "Custom Models" docs for every
+ * field.
+ */
+export interface PiProviderConfig {
+    baseUrl?: string;
+    /** 'openai-completions' | 'openai-responses' | 'anthropic-messages' | 'google-generative-ai' */
+    api?: string;
+    headers?: Record<string, string>;
+    models?: Array<{
+        id: string;
+    } & Record<string, unknown>>;
+    [key: string]: unknown;
+}
+/** What `agentOptions` accepts for the PI agent. */
+export interface PiAgentOptions {
+    /** Extra entries for PI's `~/.pi/agent/models.json`. Runs always use the
+     * `vercel-ai-gateway` provider, so this extends or overrides that provider. */
+    extraProviders?: Record<string, PiProviderConfig>;
+    /** PI thinking level: off | minimal | low | medium | high | xhigh | max. */
+    thinking?: string;
+    /** npm package (optionally versioned) to install instead of the latest PI. */
+    cliPackage?: string;
+}
+/**
+ * Generate PI's `models.json` from `agentOptions.extraProviders`, or undefined
+ * when there are none. Passed through as-is.
+ */
+export declare function generatePiModelsConfig(extraProviders: Record<string, PiProviderConfig> | undefined): string | undefined;
+/**
+ * Build the PI plugin definition.
+ *
+ * getApiKeyEnvVar() and authEnv() agree on AI_GATEWAY_API_KEY, so the host's key
+ * resolution and the sandbox env stay consistent.
+ */
+export declare function createPiCodingAgentDefinition(): AgentDefinition;
+/**
+ * Create the PI Agent. Thin wrapper over the generic orchestrator so the Agent
+ * interface (and thus registry.ts / index.ts / runner.ts) is unchanged.
+ */
+export declare function createPiCodingAgent(): Agent;
+//# sourceMappingURL=agent.d.ts.map

package/dist/lib/agents/pi-coding-agent/agent.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"agent.d.ts","sourceRoot":"","sources":["../../../../src/lib/agents/pi-coding-agent/agent.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAIH,OAAO,KAAK,EAAE,KAAK,EAAmB,MAAM,aAAa,CAAC;AAG1D,OAAO,KAAK,EAAE,eAAe,EAA2B,MAAM,uBAAuB,CAAC;AAGtF;;;;;GAKG;AACH,MAAM,WAAW,gBAAgB;IAC/B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,gGAAgG;IAChG,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC,MAAM,CAAC,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAA;KAAE,GAAG,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC;IACzD,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC;CACxB;AAED,oDAAoD;AACpD,MAAM,WAAW,cAAc;IAC7B;mFAC+E;IAC/E,cAAc,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,gBAAgB,CAAC,CAAC;IAClD,4EAA4E;IAC5E,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,8EAA8E;IAC9E,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAID;;;GAGG;AACH,wBAAgB,sBAAsB,CACpC,cAAc,EAAE,MAAM,CAAC,MAAM,EAAE,gBAAgB,CAAC,GAAG,SAAS,GAC3D,MAAM,GAAG,SAAS,CAGpB;AAyBD;;;;;GAKG;AACH,wBAAgB,6BAA6B,IAAI,eAAe,CAyD/D;AAED;;;GAGG;AACH,wBAAgB,mBAAmB,IAAI,KAAK,CAY3C"}

package/dist/lib/agents/pi-coding-agent/agent.js

@@ -0,0 +1,127 @@
+/**
+ * PI coding agent (`pi`, @earendil-works/pi-coding-agent) — host-side definition +
+ * the thin Agent wrapper.
+ *
+ * The definition is pure data/auth/config; the actual CLI invocation + transcript
+ * capture live in ./run.mjs (shipped into the sandbox by the orchestrator). PI
+ * only ever runs as a child process of that runner, inside the sandbox — its SDK
+ * is never imported into the evaluator process, so a PI failure is an agent-run
+ * failure, not an evaluator crash.
+ *
+ * PI is multi-provider, but like OpenCode it only runs through the Vercel AI
+ * Gateway here (PI's built-in `vercel-ai-gateway` provider), so there is no
+ * gateway/direct fork: a single agent, always authenticated with
+ * AI_GATEWAY_API_KEY. See {@link PiAgentOptions} for what agentOptions accepts.
+ */
+import { fileURLToPath } from 'node:url';
+import { AI_GATEWAY } from '../shared.js';
+import { runWithDefinition } from '../plugin/orchestrator.js';
+const DEFAULT_CLI_PACKAGE = '@earendil-works/pi-coding-agent';
+/**
+ * Generate PI's `models.json` from `agentOptions.extraProviders`, or undefined
+ * when there are none. Passed through as-is.
+ */
+export function generatePiModelsConfig(extraProviders) {
+    if (!extraProviders || Object.keys(extraProviders).length === 0)
+        return undefined;
+    return JSON.stringify({ providers: extraProviders }, null, 2);
+}
+/** Heredoc a config file into PI's global agent dir (an absolute `~` path writeFiles can't target). */
+function writePiAgentFile(file, content) {
+    const marker = `PI_${file.replace(/\W/g, '_').toUpperCase()}_EOF`;
+    return { viaShell: `mkdir -p ~/.pi/agent && cat > ~/.pi/agent/${file} << '${marker}'\n${content}\n${marker}` };
+}
+/**
+ * Default PI's global `defaultProjectTrust` to "always", merged into any
+ * settings.json a setup hook already wrote (an explicit trust choice wins).
+ */
+const TRUST_EVAL_WORKSPACE = {
+    viaShell: `node - << 'PI_SETTINGS_EOF'
+const fs = require('node:fs');
+const path = require('node:path');
+const file = path.join(require('node:os').homedir(), '.pi/agent/settings.json');
+let settings = {};
+try { settings = JSON.parse(fs.readFileSync(file, 'utf8')); } catch {}
+settings.defaultProjectTrust ??= 'always';
+fs.mkdirSync(path.dirname(file), { recursive: true });
+fs.writeFileSync(file, JSON.stringify(settings, null, 2));
+PI_SETTINGS_EOF`,
+};
+/**
+ * Build the PI plugin definition.
+ *
+ * getApiKeyEnvVar() and authEnv() agree on AI_GATEWAY_API_KEY, so the host's key
+ * resolution and the sandbox env stay consistent.
+ */
+export function createPiCodingAgentDefinition() {
+    return {
+        name: 'vercel-ai-gateway/pi-coding-agent',
+        displayName: 'PI Coding Agent (Vercel AI Gateway)',
+        defaultModel: 'anthropic/claude-sonnet-4.5',
+        o11yAgentName: 'vercel-ai-gateway/pi-coding-agent',
+        // PI ships no vendor-bundled skills; it only discovers caller-installed ones.
+        bundledSkillsControl: 'not-applicable',
+        // Resolve run.mjs next to this file (works in src during dev and in dist after
+        // the build copies run.mjs alongside the compiled agent.js).
+        runnerPath: fileURLToPath(new URL('./run.mjs', import.meta.url)),
+        // PI only supports the Vercel AI Gateway here, never direct provider APIs.
+        getApiKeyEnvVar() {
+            return AI_GATEWAY.apiKeyEnvVar;
+        },
+        install(options) {
+            // Project deps (retried once), then the PI CLI globally.
+            const cliPackage = options.agentOptions?.cliPackage || DEFAULT_CLI_PACKAGE;
+            return [
+                { kind: 'command', cmd: 'npm', args: ['install'], retryOnce: true, errorPrefix: 'npm install failed', errorBody: 'last10' },
+                { kind: 'command', cmd: 'npm', args: ['install', '-g', cliPackage], errorPrefix: 'PI coding agent install failed', errorBody: 'stderr' },
+            ];
+        },
+        // The gateway provider needs no config. Everything goes to PI's global agent
+        // dir, which stays out of the workspace (and so out of the captured diff).
+        configFiles(options) {
+            const piOptions = options.agentOptions;
+            // PI (>= 0.79) only loads project skills and settings (.agents/skills, .pi/*)
+            // from a trusted folder, and JSON mode cannot prompt, so it would silently
+            // skip caller-installed skills. The sandbox and its workspace are the eval's own.
+            const files = [TRUST_EVAL_WORKSPACE];
+            const models = generatePiModelsConfig(piOptions?.extraProviders);
+            if (models)
+                files.push(writePiAgentFile('models.json', models));
+            return files;
+        },
+        authEnv(options) {
+            return {
+                // No update check / install ping from inside an eval sandbox.
+                PI_SKIP_VERSION_CHECK: '1',
+                PI_TELEMETRY: '0',
+                [AI_GATEWAY.apiKeyEnvVar]: options.apiKey,
+            };
+        },
+        /**
+         * Host-threaded `--thinking` level. Rides in extra (not agentOptions) because
+         * extra is what both runner entry points ship — the orchestrator and the
+         * judge's eval-helper. null when unset.
+         */
+        runnerExtra(options) {
+            return { thinking: options.agentOptions?.thinking ?? null };
+        },
+    };
+}
+/**
+ * Create the PI Agent. Thin wrapper over the generic orchestrator so the Agent
+ * interface (and thus registry.ts / index.ts / runner.ts) is unchanged.
+ */
+export function createPiCodingAgent() {
+    const definition = createPiCodingAgentDefinition();
+    return {
+        name: definition.name,
+        displayName: definition.displayName,
+        getApiKeyEnvVar: definition.getApiKeyEnvVar,
+        getDefaultModel() {
+            return definition.defaultModel;
+        },
+        run: (fixturePath, options) => runWithDefinition(definition, fixturePath, options),
+        definition,
+    };
+}
+//# sourceMappingURL=agent.js.map

package/dist/lib/agents/pi-coding-agent/agent.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"agent.js","sourceRoot":"","sources":["../../../../src/lib/agents/pi-coding-agent/agent.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAEH,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AAIzC,OAAO,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAE1C,OAAO,EAAE,iBAAiB,EAAE,MAAM,2BAA2B,CAAC;AA4B9D,MAAM,mBAAmB,GAAG,iCAAiC,CAAC;AAE9D;;;GAGG;AACH,MAAM,UAAU,sBAAsB,CACpC,cAA4D;IAE5D,IAAI,CAAC,cAAc,IAAI,MAAM,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAClF,OAAO,IAAI,CAAC,SAAS,CAAC,EAAE,SAAS,EAAE,cAAc,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;AAChE,CAAC;AAED,uGAAuG;AACvG,SAAS,gBAAgB,CAAC,IAAY,EAAE,OAAe;IACrD,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC,WAAW,EAAE,MAAM,CAAC;IAClE,OAAO,EAAE,QAAQ,EAAE,6CAA6C,IAAI,QAAQ,MAAM,MAAM,OAAO,KAAK,MAAM,EAAE,EAAE,CAAC;AACjH,CAAC;AAED;;;GAGG;AACH,MAAM,oBAAoB,GAAe;IACvC,QAAQ,EAAE;;;;;;;;;gBASI;CACf,CAAC;AAEF;;;;;GAKG;AACH,MAAM,UAAU,6BAA6B;IAC3C,OAAO;QACL,IAAI,EAAE,mCAAmC;QACzC,WAAW,EAAE,qCAAqC;QAClD,YAAY,EAAE,6BAA6B;QAC3C,aAAa,EAAE,mCAAmC;QAClD,8EAA8E;QAC9E,oBAAoB,EAAE,gBAAgB;QACtC,+EAA+E;QAC/E,6DAA6D;QAC7D,UAAU,EAAE,aAAa,CAAC,IAAI,GAAG,CAAC,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAEhE,2EAA2E;QAC3E,eAAe;YACb,OAAO,UAAU,CAAC,YAAY,CAAC;QACjC,CAAC;QAED,OAAO,CAAC,OAAwB;YAC9B,yDAAyD;YACzD,MAAM,UAAU,GAAI,OAAO,CAAC,YAA2C,EAAE,UAAU,IAAI,mBAAmB,CAAC;YAC3G,OAAO;gBACL,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,SAAS,CAAC,EAAE,SAAS,EAAE,IAAI,EAAE,WAAW,EAAE,oBAAoB,EAAE,SAAS,EAAE,QAAQ,EAAE;gBAC3H,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,SAAS,EAAE,IAAI,EAAE,UAAU,CAAC,EAAE,WAAW,EAAE,gCAAgC,EAAE,SAAS,EAAE,QAAQ,EAAE;aACzI,CAAC;QACJ,CAAC;QAED,6EAA6E;QAC7E,2EAA2E;QAC3E,WAAW,CAAC,OAAwB;YAClC,MAAM,SAAS,GAAG,OAAO,CAAC,YAA0C,CAAC;YACrE,8EAA8E;YAC9E,2EAA2E;YAC3E,kFAAkF;YAClF,MAAM,KAAK,GAAiB,CAAC,oBAAoB,CAAC,CAAC;YACnD,MAAM,MAAM,GAAG,sBAAsB,CAAC,SAAS,EAAE,cAAc,CAAC,CAAC;YACjE,IAAI,MAAM;gBAAE,KAAK,CAAC,IAAI,CAAC,gBAAgB,CAAC,aAAa,EAAE,MAAM,CAAC,CAAC,CAAC;YAChE,OAAO,KAAK,CAAC;QACf,CAAC;QAED,OAAO,CAAC,OAAwB;YAC9B,OAAO;gBACL,8DAA8D;gBAC9D,qBAAqB,EAAE,GAAG;gBAC1B,YAAY,EAAE,GAAG;gBACjB,CAAC,UAAU,CAAC,YAAY,CAAC,EAAE,OAAO,CAAC,MAAM;aAC1C,CAAC;QACJ,CAAC;QAED;;;;WAIG;QACH,WAAW,CAAC,OAAwB;YAClC,OAAO,EAAE,QAAQ,EAAG,OAAO,CAAC,YAA2C,EAAE,QAAQ,IAAI,IAAI,EAAE,CAAC;QAC9F,CAAC;KACF,CAAC;AACJ,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,mBAAmB;IACjC,MAAM,UAAU,GAAG,6BAA6B,EAAE,CAAC;IACnD,OAAO;QACL,IAAI,EAAE,UAAU,CAAC,IAAI;QACrB,WAAW,EAAE,UAAU,CAAC,WAAW;QACnC,eAAe,EAAE,UAAU,CAAC,eAAe;QAC3C,eAAe;YACb,OAAO,UAAU,CAAC,YAAY,CAAC;QACjC,CAAC;QACD,GAAG,EAAE,CAAC,WAAW,EAAE,OAAO,EAAE,EAAE,CAAC,iBAAiB,CAAC,UAAU,EAAE,WAAW,EAAE,OAAO,CAAC;QAClF,UAAU;KACX,CAAC;AACJ,CAAC"}

package/dist/lib/agents/pi-coding-agent/run.mjs

@@ -0,0 +1,278 @@
+/**
+ * PI coding agent in-sandbox runner.
+ *
+ * This file is shipped INTO the sandbox by the orchestrator and executed there as
+ * `node __agent_eval__/run.mjs '<AgentRunInput JSON>'`. It is intentionally
+ * ZERO-DEPENDENCY (only `node:*` builtins) because the sandbox only has the
+ * fixture's own deps + the installed `pi` CLI — it cannot import anything from
+ * the @vercel/agent-eval package.
+ *
+ * Dual mode:
+ *   - runnable: invoked directly → reads argv, runs the agent, writes the result
+ *     file + prints a status line, exits 0.
+ *   - importable: `import { runAgent } from './run.mjs'` → returns a RunnerResult
+ *     (no file write, no exit). This is what the in-sandbox judge reuses.
+ *
+ * The pure helpers below are exported (not just `runAgent`) so they can be
+ * unit-tested directly — the same code the sandbox runs (see pi-coding-agent.test.ts).
+ */
+
+import { spawnSync } from 'node:child_process';
+import { mkdirSync, writeFileSync } from 'node:fs';
+import { dirname } from 'node:path';
+import { fileURLToPath } from 'node:url';
+
+/**
+ * Build the PI CLI argument list.
+ *
+ *   - `--mode json` emits every session event as JSONL on stdout (the transcript)
+ *     and exits when the agent is done. PI has no approval prompts, so there is
+ *     no "yolo" flag to pass.
+ *   - `--no-session`: the JSON stream already carries everything we persist.
+ *   - `--provider vercel-ai-gateway --model <model>` when a model override is
+ *     present. The provider is pinned because gateway ids (`anthropic/…`) would
+ *     otherwise be read as `<provider>/<model>`. On a native-default run both
+ *     are omitted — PI rejects `--provider` without `--model` — and PI picks its
+ *     own default from the only authenticated provider, the gateway.
+ *   - `--thinking <extra.thinking>` when set. HOST-threaded (agent.ts
+ *     runnerExtra) rather than read from agentOptions, so the judge's
+ *     eval-helper invocation — which ships extra but no agentOptions — gets it too.
+ *
+ * The prompt is deliberately NOT an argument: PI parses positionals starting
+ * with `@` as file attachments and `-` as flags. It is piped on stdin instead,
+ * which PI reads as the initial message (see runAgent).
+ *
+ * @param {{model?:string, extra?:Record<string,unknown>}} input
+ * @returns {string[]}
+ */
+export function buildPiCliArgs(input) {
+  const args = ['--mode', 'json', '--no-session'];
+  if (input.model) {
+    args.push('--provider', 'vercel-ai-gateway', '--model', input.model);
+  }
+  const thinking = input.extra?.thinking;
+  if (thinking) {
+    args.push('--thinking', String(thinking));
+  }
+  return args;
+}
+
+/**
+ * Events dropped from the transcript. `message_update` / `tool_execution_update`
+ * are per-token streaming deltas that each repeat the whole partial message, and
+ * `turn_end` / `agent_end` repeat every message already emitted by `message_end`
+ * — they bloat the raw stream while adding nothing the remaining events don't
+ * carry.
+ */
+const REDUNDANT_EVENT_TYPES = new Set([
+  'message_start',
+  'message_update',
+  'tool_execution_update',
+  'turn_end',
+  'agent_end',
+]);
+
+/**
+ * Parse one line of PI's `--mode json` stdout into its event, or undefined for
+ * anything that is not a JSON event.
+ *
+ * @param {string} line
+ * @returns {Record<string, any>|undefined}
+ */
+export function parsePiEventLine(line) {
+  const trimmed = line.trim();
+  if (!trimmed.startsWith('{') || !trimmed.endsWith('}')) return undefined;
+  try {
+    const event = JSON.parse(trimmed);
+    return event && typeof event.type === 'string' ? event : undefined;
+  } catch {
+    return undefined;
+  }
+}
+
+/**
+ * Whether an event is left out of the transcript (see REDUNDANT_EVENT_TYPES).
+ *
+ * @param {Record<string, any>} event
+ * @returns {boolean}
+ */
+export function isRedundantPiEvent(event) {
+  return REDUNDANT_EVENT_TYPES.has(event.type);
+}
+
+/**
+ * Serialize the kept events as the JSONL transcript.
+ *
+ * @param {Array<Record<string, any>>} events
+ * @returns {string|undefined}
+ */
+export function serializeTranscript(events) {
+  if (events.length === 0) {
+    return undefined;
+  }
+  return events.map((event) => JSON.stringify(event)).join('\n');
+}
+
+/** @param {Array<Record<string, any>>} events */
+function lastAssistantMessage(events) {
+  for (let i = events.length - 1; i >= 0; i--) {
+    const event = events[i];
+    if (event.type === 'message_end' && event.message?.role === 'assistant') {
+      return event.message;
+    }
+  }
+  return undefined;
+}
+
+/**
+ * The model PI actually used: every assistant message carries it. Reported as
+ * PI's model id — the same namespace `--model` is requested in.
+ *
+ * @param {Array<Record<string, any>>} events
+ * @returns {string|undefined}
+ */
+export function extractObservedModelFromEvents(events) {
+  const model = lastAssistantMessage(events)?.model;
+  return typeof model === 'string' && model ? model : undefined;
+}
+
+/**
+ * The run's failure, if PI reported one in-band. In JSON mode PI exits 0 even
+ * when the provider call failed (bad key, unknown model id, network error, all
+ * retries exhausted): the only signal is the final assistant message ending with
+ * stopReason 'error' / 'aborted'. Without this check such a run would be graded
+ * as a success that simply wrote no code.
+ *
+ * @param {Array<Record<string, any>>} events
+ * @returns {string|undefined}
+ */
+export function extractAgentErrorFromEvents(events) {
+  const message = lastAssistantMessage(events);
+  if (message?.stopReason === 'error' || message?.stopReason === 'aborted') {
+    return message.errorMessage || `PI stopped with reason "${message.stopReason}"`;
+  }
+  return undefined;
+}
+
+/**
+ * Run PI over the workspace at `input.cwd` and return a RunnerResult.
+ *
+ * Auth (AI_GATEWAY_API_KEY) arrives via
+ * process.env — the orchestrator sets it on the `node run.mjs` invocation, and we
+ * pass process.env straight through to the CLI. The runner never handles secrets
+ * itself.
+ *
+ * @param {import('../plugin/contract.js').AgentRunInput} input
+ * @returns {Promise<import('../plugin/contract.js').RunnerResult>}
+ */
+export async function runAgent(input) {
+  const args = buildPiCliArgs(input);
+
+  // spawnSync (not a shell string), like the other runners. The prompt goes in on
+  // stdin (see buildPiCliArgs) — which also closes the pipe: PI waits for stdin
+  // EOF before it starts. Blocking is fine — the runner has nothing else to do
+  // while the agent works. The sandbox-level timeout bounds it.
+  const res = spawnSync('pi', args, {
+    cwd: input.cwd,
+    env: process.env,
+    encoding: 'utf8',
+    maxBuffer: 64 * 1024 * 1024,
+    input: input.prompt,
+  });
+
+  const stderr = res.stderr || '';
+  // spawnSync sets status=null + error when the binary can't be spawned at all.
+  const agentExitCode = res.status == null ? -1 : res.status;
+  const spawnError = res.error;
+
+  // Split stdout into the kept events (the transcript) and anything PI printed
+  // outside the event stream.
+  const events = [];
+  let strayOutput = '';
+  for (const line of (res.stdout || '').split('\n')) {
+    const event = parsePiEventLine(line);
+    if (!event) {
+      if (line.trim()) strayOutput += line + '\n';
+    } else if (!isRedundantPiEvent(event)) {
+      events.push(event);
+    }
+  }
+
+  const transcript = serializeTranscript(events) ?? null;
+  const observedModel = extractObservedModelFromEvents(events) ?? null;
+  // stdout THEN stderr, like the other runners — with the filtered event stream
+  // standing in for the raw one.
+  const output = (transcript ? transcript + '\n' : '') + strayOutput + stderr;
+
+  const agentError = extractAgentErrorFromEvents(events);
+
+  if (spawnError || agentExitCode !== 0 || agentError) {
+    // Last 5 lines of what PI printed outside the event stream (startup errors go
+    // to stderr), else a coded fallback.
+    const errorLines = (strayOutput + stderr).trim().split('\n').slice(-5).join('\n');
+    const fallback = spawnError
+      ? `Failed to run pi: ${spawnError.message}`
+      : `PI CLI exited with code ${agentExitCode}`;
+    return {
+      ok: false,
+      output,
+      transcript,
+      observedModel,
+      error: agentError || errorLines || fallback,
+      agentExitCode,
+    };
+  }
+
+  return { ok: true, output, transcript, observedModel, error: null, agentExitCode };
+}
+
+/* ─────────────────────────── runnable (CLI) entry ─────────────────────────── */
+
+// True when this file is executed directly (`node run.mjs ...`), false when imported.
+const isMain = process.argv[1] && fileURLToPath(import.meta.url) === process.argv[1];
+
+if (isMain) {
+  // argv[2] is the AgentRunInput JSON (never contains secrets).
+  const input = JSON.parse(process.argv[2]);
+
+  // Always produce a RunnerResult, even if the runner itself throws, so the host
+  // always has a result file to read (node exit code stays 0 except on a truly
+  // unrecoverable crash before we can write).
+  let result;
+  try {
+    result = await runAgent(input);
+  } catch (e) {
+    result = {
+      ok: false,
+      output: '',
+      transcript: null,
+      observedModel: null,
+      error: e && e.message ? e.message : String(e),
+      agentExitCode: -1,
+    };
+  }
+
+  // Source of truth: the result file the host reads back via sandbox.readFile.
+  try {
+    mkdirSync(dirname(input.resultPath), { recursive: true });
+    writeFileSync(input.resultPath, JSON.stringify(result));
+  } catch {
+    // If the file can't be written, the host falls back to the marker line below.
+  }
+
+  // Fallback channel: a compact status line (no transcript — it can be huge).
+  process.stdout.write(
+    '__AGENT_RESULT__ ' +
+      JSON.stringify({
+        ok: result.ok,
+        observedModel: result.observedModel,
+        error: result.error,
+        agentExitCode: result.agentExitCode,
+      }) +
+      '\n'
+  );
+
+  // Exit 0: "the runner ran". Agent success/failure is conveyed via result.ok, not
+  // the node exit code (the host distinguishes the two).
+  process.exit(0);
+}

package/dist/lib/agents/plugin/contract.d.ts

@@ -19,6 +19,7 @@
  * host. See the migration notes in the PR for the full rationale.
  */
 import type { ModelTier, RunnableExperimentConfig } from '../../types.js';
+import type { TokenUsage } from '../../o11y/types.js';
 import type { AgentRunOptions } from '../types.js';
 /**
  * One install action the orchestrator runs in the sandbox before invocation.
@@ -38,6 +39,13 @@ export interface InstallStep {
     script?: string;
     /** Retry once on a non-zero exit before failing (the project `npm install`). */
     retryOnce?: boolean;
+    /**
+     * What the step installs. `'project'` marks the fixture's own dependencies:
+     * a separate verifier re-runs only those steps (plus the agent steps when a
+     * judge needs the CLI). When no step in a definition sets a scope, the
+     * verifier runs every step.
+     */
+    scope?: 'project' | 'agent';
     /** Prefix of the thrown Error on final failure, e.g. 'Claude Code install failed'. */
     errorPrefix: string;
     /**
@@ -60,6 +68,22 @@ export interface ConfigFile {
     /** Exact shell to run instead, e.g. `mkdir -p ~/.codex && cat > ~/.codex/... << 'EOF' ...`. */
     viaShell?: string;
 }
+/**
+ * A command that prints the agent CLI's version: `cmd` with `args` or, for kind
+ * 'shell', a bash `script`. It is given `timeoutMs` (default 30 seconds) to
+ * finish; after that it is abandoned, killed in the sandbox where coreutils
+ * `timeout` is available, and the version is left unset.
+ */
+export type VersionCommand = {
+    kind: 'command';
+    cmd: string;
+    args?: string[];
+    timeoutMs?: number;
+} | {
+    kind: 'shell';
+    script: string;
+    timeoutMs?: number;
+};
 /**
  * Host-side plugin description for one agent. Pure data + tiny pure functions —
  * no sandbox calls, no CLI invocation.
@@ -93,6 +117,14 @@ export interface AgentDefinition {
     install(options: AgentRunOptions): InstallStep[];
     /** Config files written before invocation (codex TOML, opencode.json). [] for most. */
     configFiles(options: AgentRunOptions): ConfigFile[];
+    /**
+     * OPTIONAL command that prints the installed CLI's version. The orchestrator
+     * runs it once after install and records its trimmed stdout as
+     * `provenance.agentCliVersion`. A non-zero exit, empty output, or error leaves
+     * the version unset and never fails the run. Omit it (or return undefined)
+     * when the CLI has no version command.
+     */
+    versionCommand?(options: AgentRunOptions): VersionCommand | undefined;
     /**
      * The agent-specific auth/env handed to the run.mjs process (merged with the
      * neutral-workspace env by the orchestrator). This is the ONLY place provider
@@ -184,5 +216,13 @@ export interface RunnerResult {
      * repairs dropping to zero means the upstream CLI is fixed).
      */
     modelRepair?: string;
+    /**
+     * OPTIONAL token usage the CLI reported somewhere other than the transcript.
+     * fx prints its usage only in `fx ask --json`, while the transcript is the
+     * saved session, which records context size rather than consumed tokens.
+     * When set, it takes precedence over usage read from the transcript; omit it
+     * to have usage read from the transcript as usual.
+     */
+    usage?: TokenUsage;
 }
 //# sourceMappingURL=contract.d.ts.map