prose-qa 0.1.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/LICENSE +21 -0
- package/README.md +570 -0
- package/dist/agent/bash.d.ts +52 -0
- package/dist/agent/bash.d.ts.map +1 -0
- package/dist/agent/bash.js +186 -0
- package/dist/agent/bash.js.map +1 -0
- package/dist/agent/bash.test.d.ts +2 -0
- package/dist/agent/bash.test.d.ts.map +1 -0
- package/dist/agent/bash.test.js +70 -0
- package/dist/agent/bash.test.js.map +1 -0
- package/dist/agent/llm-model.d.ts +5 -0
- package/dist/agent/llm-model.d.ts.map +1 -0
- package/dist/agent/llm-model.js +29 -0
- package/dist/agent/llm-model.js.map +1 -0
- package/dist/agent/llm-model.test.d.ts +2 -0
- package/dist/agent/llm-model.test.d.ts.map +1 -0
- package/dist/agent/llm-model.test.js +28 -0
- package/dist/agent/llm-model.test.js.map +1 -0
- package/dist/agent/prompt.d.ts +17 -0
- package/dist/agent/prompt.d.ts.map +1 -0
- package/dist/agent/prompt.js +97 -0
- package/dist/agent/prompt.js.map +1 -0
- package/dist/agent/prompt.test.d.ts +2 -0
- package/dist/agent/prompt.test.d.ts.map +1 -0
- package/dist/agent/prompt.test.js +124 -0
- package/dist/agent/prompt.test.js.map +1 -0
- package/dist/agent/provider-options.d.ts +8 -0
- package/dist/agent/provider-options.d.ts.map +1 -0
- package/dist/agent/provider-options.js +115 -0
- package/dist/agent/provider-options.js.map +1 -0
- package/dist/agent/provider-options.test.d.ts +2 -0
- package/dist/agent/provider-options.test.d.ts.map +1 -0
- package/dist/agent/provider-options.test.js +114 -0
- package/dist/agent/provider-options.test.js.map +1 -0
- package/dist/agent/runner.d.ts +27 -0
- package/dist/agent/runner.d.ts.map +1 -0
- package/dist/agent/runner.js +291 -0
- package/dist/agent/runner.js.map +1 -0
- package/dist/agent/verdict-retry-prompt.d.ts +3 -0
- package/dist/agent/verdict-retry-prompt.d.ts.map +1 -0
- package/dist/agent/verdict-retry-prompt.js +18 -0
- package/dist/agent/verdict-retry-prompt.js.map +1 -0
- package/dist/agent/verdict-retry-prompt.test.d.ts +2 -0
- package/dist/agent/verdict-retry-prompt.test.d.ts.map +1 -0
- package/dist/agent/verdict-retry-prompt.test.js +25 -0
- package/dist/agent/verdict-retry-prompt.test.js.map +1 -0
- package/dist/agent/verdict.d.ts +31 -0
- package/dist/agent/verdict.d.ts.map +1 -0
- package/dist/agent/verdict.js +123 -0
- package/dist/agent/verdict.js.map +1 -0
- package/dist/agent/verdict.test.d.ts +2 -0
- package/dist/agent/verdict.test.d.ts.map +1 -0
- package/dist/agent/verdict.test.js +156 -0
- package/dist/agent/verdict.test.js.map +1 -0
- package/dist/analyze/build-context.d.ts +58 -0
- package/dist/analyze/build-context.d.ts.map +1 -0
- package/dist/analyze/build-context.js +141 -0
- package/dist/analyze/build-context.js.map +1 -0
- package/dist/analyze/build-context.test.d.ts +2 -0
- package/dist/analyze/build-context.test.d.ts.map +1 -0
- package/dist/analyze/build-context.test.js +118 -0
- package/dist/analyze/build-context.test.js.map +1 -0
- package/dist/analyze/compare-runs.d.ts +49 -0
- package/dist/analyze/compare-runs.d.ts.map +1 -0
- package/dist/analyze/compare-runs.js +214 -0
- package/dist/analyze/compare-runs.js.map +1 -0
- package/dist/analyze/compare-runs.test.d.ts +2 -0
- package/dist/analyze/compare-runs.test.d.ts.map +1 -0
- package/dist/analyze/compare-runs.test.js +139 -0
- package/dist/analyze/compare-runs.test.js.map +1 -0
- package/dist/analyze/diff-hunks.d.ts +16 -0
- package/dist/analyze/diff-hunks.d.ts.map +1 -0
- package/dist/analyze/diff-hunks.js +287 -0
- package/dist/analyze/diff-hunks.js.map +1 -0
- package/dist/analyze/diff-hunks.test.d.ts +2 -0
- package/dist/analyze/diff-hunks.test.d.ts.map +1 -0
- package/dist/analyze/diff-hunks.test.js +54 -0
- package/dist/analyze/diff-hunks.test.js.map +1 -0
- package/dist/analyze/hunk-editor.d.ts +8 -0
- package/dist/analyze/hunk-editor.d.ts.map +1 -0
- package/dist/analyze/hunk-editor.js +129 -0
- package/dist/analyze/hunk-editor.js.map +1 -0
- package/dist/analyze/hunk-editor.test.d.ts +2 -0
- package/dist/analyze/hunk-editor.test.d.ts.map +1 -0
- package/dist/analyze/hunk-editor.test.js +48 -0
- package/dist/analyze/hunk-editor.test.js.map +1 -0
- package/dist/analyze/index.d.ts +23 -0
- package/dist/analyze/index.d.ts.map +1 -0
- package/dist/analyze/index.js +122 -0
- package/dist/analyze/index.js.map +1 -0
- package/dist/analyze/llm-fix.d.ts +11 -0
- package/dist/analyze/llm-fix.d.ts.map +1 -0
- package/dist/analyze/llm-fix.js +76 -0
- package/dist/analyze/llm-fix.js.map +1 -0
- package/dist/analyze/parse-proposal.d.ts +41 -0
- package/dist/analyze/parse-proposal.d.ts.map +1 -0
- package/dist/analyze/parse-proposal.js +53 -0
- package/dist/analyze/parse-proposal.js.map +1 -0
- package/dist/analyze/parse-proposal.test.d.ts +2 -0
- package/dist/analyze/parse-proposal.test.d.ts.map +1 -0
- package/dist/analyze/parse-proposal.test.js +40 -0
- package/dist/analyze/parse-proposal.test.js.map +1 -0
- package/dist/analyze/repl.d.ts +28 -0
- package/dist/analyze/repl.d.ts.map +1 -0
- package/dist/analyze/repl.js +284 -0
- package/dist/analyze/repl.js.map +1 -0
- package/dist/analyze/repl.test.d.ts +2 -0
- package/dist/analyze/repl.test.d.ts.map +1 -0
- package/dist/analyze/repl.test.js +101 -0
- package/dist/analyze/repl.test.js.map +1 -0
- package/dist/analyze/suggest.d.ts +5 -0
- package/dist/analyze/suggest.d.ts.map +1 -0
- package/dist/analyze/suggest.js +75 -0
- package/dist/analyze/suggest.js.map +1 -0
- package/dist/analyze/suggest.test.d.ts +2 -0
- package/dist/analyze/suggest.test.d.ts.map +1 -0
- package/dist/analyze/suggest.test.js +53 -0
- package/dist/analyze/suggest.test.js.map +1 -0
- package/dist/analyze/validate-markdown.d.ts +3 -0
- package/dist/analyze/validate-markdown.d.ts.map +1 -0
- package/dist/analyze/validate-markdown.js +25 -0
- package/dist/analyze/validate-markdown.js.map +1 -0
- package/dist/artifacts/policy.d.ts +9 -0
- package/dist/artifacts/policy.d.ts.map +1 -0
- package/dist/artifacts/policy.js +46 -0
- package/dist/artifacts/policy.js.map +1 -0
- package/dist/artifacts/policy.test.d.ts +2 -0
- package/dist/artifacts/policy.test.d.ts.map +1 -0
- package/dist/artifacts/policy.test.js +73 -0
- package/dist/artifacts/policy.test.js.map +1 -0
- package/dist/auth/resolve.d.ts +22 -0
- package/dist/auth/resolve.d.ts.map +1 -0
- package/dist/auth/resolve.js +148 -0
- package/dist/auth/resolve.js.map +1 -0
- package/dist/auth/store.d.ts +23 -0
- package/dist/auth/store.d.ts.map +1 -0
- package/dist/auth/store.js +103 -0
- package/dist/auth/store.js.map +1 -0
- package/dist/cache/generate.d.ts +8 -0
- package/dist/cache/generate.d.ts.map +1 -0
- package/dist/cache/generate.js +61 -0
- package/dist/cache/generate.js.map +1 -0
- package/dist/cache/hash.d.ts +5 -0
- package/dist/cache/hash.d.ts.map +1 -0
- package/dist/cache/hash.js +21 -0
- package/dist/cache/hash.js.map +1 -0
- package/dist/cache/hash.test.d.ts +2 -0
- package/dist/cache/hash.test.d.ts.map +1 -0
- package/dist/cache/hash.test.js +42 -0
- package/dist/cache/hash.test.js.map +1 -0
- package/dist/cache/resolve.d.ts +5 -0
- package/dist/cache/resolve.d.ts.map +1 -0
- package/dist/cache/resolve.js +8 -0
- package/dist/cache/resolve.js.map +1 -0
- package/dist/cache/store.d.ts +20 -0
- package/dist/cache/store.d.ts.map +1 -0
- package/dist/cache/store.js +90 -0
- package/dist/cache/store.js.map +1 -0
- package/dist/cache/store.test.d.ts +2 -0
- package/dist/cache/store.test.d.ts.map +1 -0
- package/dist/cache/store.test.js +101 -0
- package/dist/cache/store.test.js.map +1 -0
- package/dist/cli/analyze.d.ts +21 -0
- package/dist/cli/analyze.d.ts.map +1 -0
- package/dist/cli/analyze.js +148 -0
- package/dist/cli/analyze.js.map +1 -0
- package/dist/cli/concurrency.d.ts +17 -0
- package/dist/cli/concurrency.d.ts.map +1 -0
- package/dist/cli/concurrency.js +56 -0
- package/dist/cli/concurrency.js.map +1 -0
- package/dist/cli/concurrency.test.d.ts +2 -0
- package/dist/cli/concurrency.test.d.ts.map +1 -0
- package/dist/cli/concurrency.test.js +74 -0
- package/dist/cli/concurrency.test.js.map +1 -0
- package/dist/cli/config.d.ts +2 -0
- package/dist/cli/config.d.ts.map +1 -0
- package/dist/cli/config.js +14 -0
- package/dist/cli/config.js.map +1 -0
- package/dist/cli/help.d.ts +23 -0
- package/dist/cli/help.d.ts.map +1 -0
- package/dist/cli/help.js +458 -0
- package/dist/cli/help.js.map +1 -0
- package/dist/cli/help.test.d.ts +2 -0
- package/dist/cli/help.test.d.ts.map +1 -0
- package/dist/cli/help.test.js +41 -0
- package/dist/cli/help.test.js.map +1 -0
- package/dist/cli/index.d.ts +3 -0
- package/dist/cli/index.d.ts.map +1 -0
- package/dist/cli/index.js +300 -0
- package/dist/cli/index.js.map +1 -0
- package/dist/cli/mcp.d.ts +6 -0
- package/dist/cli/mcp.d.ts.map +1 -0
- package/dist/cli/mcp.js +17 -0
- package/dist/cli/mcp.js.map +1 -0
- package/dist/cli/record.d.ts +27 -0
- package/dist/cli/record.d.ts.map +1 -0
- package/dist/cli/record.js +244 -0
- package/dist/cli/record.js.map +1 -0
- package/dist/cli/run.d.ts +11 -0
- package/dist/cli/run.d.ts.map +1 -0
- package/dist/cli/run.js +676 -0
- package/dist/cli/run.js.map +1 -0
- package/dist/cli/subprocess.d.ts +19 -0
- package/dist/cli/subprocess.d.ts.map +1 -0
- package/dist/cli/subprocess.js +142 -0
- package/dist/cli/subprocess.js.map +1 -0
- package/dist/cli/subprocess.test.d.ts +2 -0
- package/dist/cli/subprocess.test.d.ts.map +1 -0
- package/dist/cli/subprocess.test.js +76 -0
- package/dist/cli/subprocess.test.js.map +1 -0
- package/dist/cli/tags.d.ts +5 -0
- package/dist/cli/tags.d.ts.map +1 -0
- package/dist/cli/tags.js +33 -0
- package/dist/cli/tags.js.map +1 -0
- package/dist/cli/tags.test.d.ts +2 -0
- package/dist/cli/tags.test.d.ts.map +1 -0
- package/dist/cli/tags.test.js +31 -0
- package/dist/cli/tags.test.js.map +1 -0
- package/dist/config/env-vars.d.ts +2 -0
- package/dist/config/env-vars.d.ts.map +1 -0
- package/dist/config/env-vars.js +14 -0
- package/dist/config/env-vars.js.map +1 -0
- package/dist/config/env.d.ts +2 -0
- package/dist/config/env.d.ts.map +1 -0
- package/dist/config/env.js +9 -0
- package/dist/config/env.js.map +1 -0
- package/dist/config/lightpanda.d.ts +6 -0
- package/dist/config/lightpanda.d.ts.map +1 -0
- package/dist/config/lightpanda.js +38 -0
- package/dist/config/lightpanda.js.map +1 -0
- package/dist/config/lightpanda.test.d.ts +2 -0
- package/dist/config/lightpanda.test.d.ts.map +1 -0
- package/dist/config/lightpanda.test.js +46 -0
- package/dist/config/lightpanda.test.js.map +1 -0
- package/dist/config/load.d.ts +22 -0
- package/dist/config/load.d.ts.map +1 -0
- package/dist/config/load.js +242 -0
- package/dist/config/load.js.map +1 -0
- package/dist/config/load.test.d.ts +2 -0
- package/dist/config/load.test.d.ts.map +1 -0
- package/dist/config/load.test.js +86 -0
- package/dist/config/load.test.js.map +1 -0
- package/dist/config/set.d.ts +8 -0
- package/dist/config/set.d.ts.map +1 -0
- package/dist/config/set.js +93 -0
- package/dist/config/set.js.map +1 -0
- package/dist/config/set.test.d.ts +2 -0
- package/dist/config/set.test.d.ts.map +1 -0
- package/dist/config/set.test.js +98 -0
- package/dist/config/set.test.js.map +1 -0
- package/dist/healing/classify.d.ts +15 -0
- package/dist/healing/classify.d.ts.map +1 -0
- package/dist/healing/classify.js +209 -0
- package/dist/healing/classify.js.map +1 -0
- package/dist/healing/classify.test.d.ts +2 -0
- package/dist/healing/classify.test.d.ts.map +1 -0
- package/dist/healing/classify.test.js +167 -0
- package/dist/healing/classify.test.js.map +1 -0
- package/dist/healing/recovery-prompt.d.ts +3 -0
- package/dist/healing/recovery-prompt.d.ts.map +1 -0
- package/dist/healing/recovery-prompt.js +22 -0
- package/dist/healing/recovery-prompt.js.map +1 -0
- package/dist/mcp/inline-scenario.d.ts +13 -0
- package/dist/mcp/inline-scenario.d.ts.map +1 -0
- package/dist/mcp/inline-scenario.js +23 -0
- package/dist/mcp/inline-scenario.js.map +1 -0
- package/dist/mcp/server.d.ts +4 -0
- package/dist/mcp/server.d.ts.map +1 -0
- package/dist/mcp/server.js +186 -0
- package/dist/mcp/server.js.map +1 -0
- package/dist/mcp/skill.d.ts +5 -0
- package/dist/mcp/skill.d.ts.map +1 -0
- package/dist/mcp/skill.js +38 -0
- package/dist/mcp/skill.js.map +1 -0
- package/dist/mcp/skill.test.d.ts +2 -0
- package/dist/mcp/skill.test.d.ts.map +1 -0
- package/dist/mcp/skill.test.js +18 -0
- package/dist/mcp/skill.test.js.map +1 -0
- package/dist/paths.d.ts +12 -0
- package/dist/paths.d.ts.map +1 -0
- package/dist/paths.js +61 -0
- package/dist/paths.js.map +1 -0
- package/dist/prompt/load.d.ts +4 -0
- package/dist/prompt/load.d.ts.map +1 -0
- package/dist/prompt/load.js +19 -0
- package/dist/prompt/load.js.map +1 -0
- package/dist/recorder/bridge-process.d.ts +14 -0
- package/dist/recorder/bridge-process.d.ts.map +1 -0
- package/dist/recorder/bridge-process.js +133 -0
- package/dist/recorder/bridge-process.js.map +1 -0
- package/dist/recorder/bridge-process.test.d.ts +2 -0
- package/dist/recorder/bridge-process.test.d.ts.map +1 -0
- package/dist/recorder/bridge-process.test.js +36 -0
- package/dist/recorder/bridge-process.test.js.map +1 -0
- package/dist/recorder/bridge-worker.d.ts +2 -0
- package/dist/recorder/bridge-worker.d.ts.map +1 -0
- package/dist/recorder/bridge-worker.js +76 -0
- package/dist/recorder/bridge-worker.js.map +1 -0
- package/dist/recorder/bridge.d.ts +12 -0
- package/dist/recorder/bridge.d.ts.map +1 -0
- package/dist/recorder/bridge.js +61 -0
- package/dist/recorder/bridge.js.map +1 -0
- package/dist/recorder/bridge.test.d.ts +2 -0
- package/dist/recorder/bridge.test.d.ts.map +1 -0
- package/dist/recorder/bridge.test.js +21 -0
- package/dist/recorder/bridge.test.js.map +1 -0
- package/dist/recorder/enrich-event.d.ts +31 -0
- package/dist/recorder/enrich-event.d.ts.map +1 -0
- package/dist/recorder/enrich-event.js +91 -0
- package/dist/recorder/enrich-event.js.map +1 -0
- package/dist/recorder/events.d.ts +11 -0
- package/dist/recorder/events.d.ts.map +1 -0
- package/dist/recorder/events.js +42 -0
- package/dist/recorder/events.js.map +1 -0
- package/dist/recorder/events.test.d.ts +2 -0
- package/dist/recorder/events.test.d.ts.map +1 -0
- package/dist/recorder/events.test.js +40 -0
- package/dist/recorder/events.test.js.map +1 -0
- package/dist/recorder/generate-scenario.d.ts +16 -0
- package/dist/recorder/generate-scenario.d.ts.map +1 -0
- package/dist/recorder/generate-scenario.js +78 -0
- package/dist/recorder/generate-scenario.js.map +1 -0
- package/dist/recorder/in-page-helpers.d.ts +6 -0
- package/dist/recorder/in-page-helpers.d.ts.map +1 -0
- package/dist/recorder/in-page-helpers.js +238 -0
- package/dist/recorder/in-page-helpers.js.map +1 -0
- package/dist/recorder/in-page-helpers.test.d.ts +2 -0
- package/dist/recorder/in-page-helpers.test.d.ts.map +1 -0
- package/dist/recorder/in-page-helpers.test.js +186 -0
- package/dist/recorder/in-page-helpers.test.js.map +1 -0
- package/dist/recorder/page-script.d.ts +7 -0
- package/dist/recorder/page-script.d.ts.map +1 -0
- package/dist/recorder/page-script.js +132 -0
- package/dist/recorder/page-script.js.map +1 -0
- package/dist/recorder/redact.d.ts +8 -0
- package/dist/recorder/redact.d.ts.map +1 -0
- package/dist/recorder/redact.js +26 -0
- package/dist/recorder/redact.js.map +1 -0
- package/dist/recorder/redact.test.d.ts +2 -0
- package/dist/recorder/redact.test.d.ts.map +1 -0
- package/dist/recorder/redact.test.js +27 -0
- package/dist/recorder/redact.test.js.map +1 -0
- package/dist/recorder/session.d.ts +8 -0
- package/dist/recorder/session.d.ts.map +1 -0
- package/dist/recorder/session.js +28 -0
- package/dist/recorder/session.js.map +1 -0
- package/dist/recorder/snapshot-match.d.ts +22 -0
- package/dist/recorder/snapshot-match.d.ts.map +1 -0
- package/dist/recorder/snapshot-match.js +102 -0
- package/dist/recorder/snapshot-match.js.map +1 -0
- package/dist/recorder/snapshot-match.test.d.ts +2 -0
- package/dist/recorder/snapshot-match.test.d.ts.map +1 -0
- package/dist/recorder/snapshot-match.test.js +34 -0
- package/dist/recorder/snapshot-match.test.js.map +1 -0
- package/dist/redact/env-secrets.d.ts +14 -0
- package/dist/redact/env-secrets.d.ts.map +1 -0
- package/dist/redact/env-secrets.js +86 -0
- package/dist/redact/env-secrets.js.map +1 -0
- package/dist/redact/env-secrets.test.d.ts +2 -0
- package/dist/redact/env-secrets.test.d.ts.map +1 -0
- package/dist/redact/env-secrets.test.js +103 -0
- package/dist/redact/env-secrets.test.js.map +1 -0
- package/dist/reporter/export.d.ts +14 -0
- package/dist/reporter/export.d.ts.map +1 -0
- package/dist/reporter/export.js +53 -0
- package/dist/reporter/export.js.map +1 -0
- package/dist/reporter/export.test.d.ts +2 -0
- package/dist/reporter/export.test.d.ts.map +1 -0
- package/dist/reporter/export.test.js +100 -0
- package/dist/reporter/export.test.js.map +1 -0
- package/dist/reporter/index.d.ts +11 -0
- package/dist/reporter/index.d.ts.map +1 -0
- package/dist/reporter/index.js +161 -0
- package/dist/reporter/index.js.map +1 -0
- package/dist/reporter/index.test.d.ts +2 -0
- package/dist/reporter/index.test.d.ts.map +1 -0
- package/dist/reporter/index.test.js +61 -0
- package/dist/reporter/index.test.js.map +1 -0
- package/dist/scenarios/globs.d.ts +15 -0
- package/dist/scenarios/globs.d.ts.map +1 -0
- package/dist/scenarios/globs.js +48 -0
- package/dist/scenarios/globs.js.map +1 -0
- package/dist/scenarios/globs.test.d.ts +2 -0
- package/dist/scenarios/globs.test.d.ts.map +1 -0
- package/dist/scenarios/globs.test.js +53 -0
- package/dist/scenarios/globs.test.js.map +1 -0
- package/dist/scenarios/parser.d.ts +15 -0
- package/dist/scenarios/parser.d.ts.map +1 -0
- package/dist/scenarios/parser.js +278 -0
- package/dist/scenarios/parser.js.map +1 -0
- package/dist/scenarios/parser.test.d.ts +2 -0
- package/dist/scenarios/parser.test.d.ts.map +1 -0
- package/dist/scenarios/parser.test.js +373 -0
- package/dist/scenarios/parser.test.js.map +1 -0
- package/dist/skills/loader.d.ts +10 -0
- package/dist/skills/loader.d.ts.map +1 -0
- package/dist/skills/loader.js +98 -0
- package/dist/skills/loader.js.map +1 -0
- package/dist/types/config.d.ts +131 -0
- package/dist/types/config.d.ts.map +1 -0
- package/dist/types/config.js +2 -0
- package/dist/types/config.js.map +1 -0
- package/dist/types/recorder.d.ts +71 -0
- package/dist/types/recorder.d.ts.map +1 -0
- package/dist/types/recorder.js +2 -0
- package/dist/types/recorder.js.map +1 -0
- package/dist/types/scenario.d.ts +41 -0
- package/dist/types/scenario.d.ts.map +1 -0
- package/dist/types/scenario.js +2 -0
- package/dist/types/scenario.js.map +1 -0
- package/dist/types/skill.d.ts +20 -0
- package/dist/types/skill.d.ts.map +1 -0
- package/dist/types/skill.js +13 -0
- package/dist/types/skill.js.map +1 -0
- package/dist/types/verdict.d.ts +82 -0
- package/dist/types/verdict.d.ts.map +1 -0
- package/dist/types/verdict.js +13 -0
- package/dist/types/verdict.js.map +1 -0
- package/package.json +75 -0
- package/pqa.config.ts +82 -0
- package/prompt/ANALYZE-FLAKY.md +62 -0
- package/prompt/ANALYZE.md +110 -0
- package/prompt/CACHE-HINTS.md +49 -0
- package/prompt/RECORD.md +114 -0
- package/prompt/SYSTEM.md +118 -0
- package/skills/agent-browser/SKILL.md +2438 -0
- package/skills/create-pqa-scenario/SKILL.md +273 -0
|
@@ -0,0 +1,8 @@
|
|
|
1
|
+
import type { ProviderOptions } from "@ai-sdk/provider-utils";
|
|
2
|
+
import type { PqaConfig } from "../types/config.js";
|
|
3
|
+
/**
|
|
4
|
+
* Provider-specific options for extended thinking / reasoning.
|
|
5
|
+
* Honors `config.llm.thinking.enabled` across Anthropic, OpenAI, Fireworks, Google, OpenRouter, and Ollama.
|
|
6
|
+
*/
|
|
7
|
+
export declare function buildProviderOptions(config: PqaConfig): ProviderOptions | undefined;
|
|
8
|
+
//# sourceMappingURL=provider-options.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"provider-options.d.ts","sourceRoot":"","sources":["../../src/agent/provider-options.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AAC9D,OAAO,KAAK,EAAsB,SAAS,EAAE,MAAM,oBAAoB,CAAC;AA0DxE;;;GAGG;AACH,wBAAgB,oBAAoB,CAClC,MAAM,EAAE,SAAS,GAChB,eAAe,GAAG,SAAS,CA2D7B"}
|
|
@@ -0,0 +1,115 @@
|
|
|
1
|
+
const DEFAULT_THINKING_BUDGET = 10_000;
|
|
2
|
+
function thinkingBudget(config) {
|
|
3
|
+
return config.llm.thinking?.budgetTokens ?? DEFAULT_THINKING_BUDGET;
|
|
4
|
+
}
|
|
5
|
+
function isThinkingEnabled(config) {
|
|
6
|
+
return config.llm.thinking?.enabled === true;
|
|
7
|
+
}
|
|
8
|
+
/** Derive OpenAI reasoning effort from token budget when `reasoningEffort` is unset. */
|
|
9
|
+
function reasoningEffortFromBudget(budgetTokens) {
|
|
10
|
+
if (budgetTokens <= 2_000)
|
|
11
|
+
return "minimal";
|
|
12
|
+
if (budgetTokens <= 5_000)
|
|
13
|
+
return "low";
|
|
14
|
+
if (budgetTokens <= 10_000)
|
|
15
|
+
return "medium";
|
|
16
|
+
if (budgetTokens <= 20_000)
|
|
17
|
+
return "high";
|
|
18
|
+
return "xhigh";
|
|
19
|
+
}
|
|
20
|
+
function resolveOpenAIReasoningEffort(config) {
|
|
21
|
+
const explicit = config.llm.thinking?.reasoningEffort;
|
|
22
|
+
if (explicit != null)
|
|
23
|
+
return explicit;
|
|
24
|
+
return reasoningEffortFromBudget(thinkingBudget(config));
|
|
25
|
+
}
|
|
26
|
+
function geminiThinkingLevel(reasoningEffort, budgetTokens) {
|
|
27
|
+
if (reasoningEffort != null && reasoningEffort !== "none") {
|
|
28
|
+
if (reasoningEffort === "xhigh")
|
|
29
|
+
return "high";
|
|
30
|
+
if (reasoningEffort === "minimal")
|
|
31
|
+
return "minimal";
|
|
32
|
+
return reasoningEffort;
|
|
33
|
+
}
|
|
34
|
+
if (budgetTokens <= 2_000)
|
|
35
|
+
return "minimal";
|
|
36
|
+
if (budgetTokens <= 5_000)
|
|
37
|
+
return "low";
|
|
38
|
+
if (budgetTokens <= 10_000)
|
|
39
|
+
return "medium";
|
|
40
|
+
return "high";
|
|
41
|
+
}
|
|
42
|
+
/** Map config effort to Anthropic `effort` (none/minimal are omitted). */
|
|
43
|
+
function anthropicEffort(reasoningEffort) {
|
|
44
|
+
if (reasoningEffort == null || reasoningEffort === "none") {
|
|
45
|
+
return undefined;
|
|
46
|
+
}
|
|
47
|
+
if (reasoningEffort === "minimal")
|
|
48
|
+
return "low";
|
|
49
|
+
if (reasoningEffort === "xhigh")
|
|
50
|
+
return "xhigh";
|
|
51
|
+
return reasoningEffort;
|
|
52
|
+
}
|
|
53
|
+
/**
|
|
54
|
+
* Provider-specific options for extended thinking / reasoning.
|
|
55
|
+
* Honors `config.llm.thinking.enabled` across Anthropic, OpenAI, Fireworks, Google, OpenRouter, and Ollama.
|
|
56
|
+
*/
|
|
57
|
+
export function buildProviderOptions(config) {
|
|
58
|
+
if (!isThinkingEnabled(config)) {
|
|
59
|
+
if (config.llm.provider === "anthropic") {
|
|
60
|
+
return { anthropic: { disableParallelToolUse: true } };
|
|
61
|
+
}
|
|
62
|
+
return undefined;
|
|
63
|
+
}
|
|
64
|
+
const budget = thinkingBudget(config);
|
|
65
|
+
const reasoningEffort = config.llm.thinking?.reasoningEffort;
|
|
66
|
+
switch (config.llm.provider) {
|
|
67
|
+
case "anthropic": {
|
|
68
|
+
const effort = anthropicEffort(reasoningEffort);
|
|
69
|
+
return {
|
|
70
|
+
anthropic: {
|
|
71
|
+
disableParallelToolUse: true,
|
|
72
|
+
thinking: { type: "enabled", budgetTokens: budget },
|
|
73
|
+
...(effort && { effort }),
|
|
74
|
+
},
|
|
75
|
+
};
|
|
76
|
+
}
|
|
77
|
+
case "openai":
|
|
78
|
+
return {
|
|
79
|
+
openai: {
|
|
80
|
+
reasoningEffort: resolveOpenAIReasoningEffort(config),
|
|
81
|
+
},
|
|
82
|
+
};
|
|
83
|
+
case "fireworks":
|
|
84
|
+
return {
|
|
85
|
+
fireworks: {
|
|
86
|
+
thinking: { type: "enabled", budgetTokens: budget },
|
|
87
|
+
},
|
|
88
|
+
};
|
|
89
|
+
case "google":
|
|
90
|
+
return {
|
|
91
|
+
google: {
|
|
92
|
+
thinkingConfig: {
|
|
93
|
+
includeThoughts: true,
|
|
94
|
+
thinkingBudget: budget,
|
|
95
|
+
thinkingLevel: geminiThinkingLevel(reasoningEffort, budget),
|
|
96
|
+
},
|
|
97
|
+
},
|
|
98
|
+
};
|
|
99
|
+
case "ollama":
|
|
100
|
+
return {
|
|
101
|
+
ollama: { think: true },
|
|
102
|
+
};
|
|
103
|
+
case "openrouter":
|
|
104
|
+
return {
|
|
105
|
+
openrouter: {
|
|
106
|
+
reasoning: reasoningEffort
|
|
107
|
+
? { effort: reasoningEffort }
|
|
108
|
+
: { max_tokens: budget },
|
|
109
|
+
},
|
|
110
|
+
};
|
|
111
|
+
default:
|
|
112
|
+
return undefined;
|
|
113
|
+
}
|
|
114
|
+
}
|
|
115
|
+
//# sourceMappingURL=provider-options.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"provider-options.js","sourceRoot":"","sources":["../../src/agent/provider-options.ts"],"names":[],"mappings":"AAGA,MAAM,uBAAuB,GAAG,MAAM,CAAC;AAIvC,SAAS,cAAc,CAAC,MAAiB;IACvC,OAAO,MAAM,CAAC,GAAG,CAAC,QAAQ,EAAE,YAAY,IAAI,uBAAuB,CAAC;AACtE,CAAC;AAED,SAAS,iBAAiB,CAAC,MAAiB;IAC1C,OAAO,MAAM,CAAC,GAAG,CAAC,QAAQ,EAAE,OAAO,KAAK,IAAI,CAAC;AAC/C,CAAC;AAED,wFAAwF;AACxF,SAAS,yBAAyB,CAAC,YAAoB;IACrD,IAAI,YAAY,IAAI,KAAK;QAAE,OAAO,SAAS,CAAC;IAC5C,IAAI,YAAY,IAAI,KAAK;QAAE,OAAO,KAAK,CAAC;IACxC,IAAI,YAAY,IAAI,MAAM;QAAE,OAAO,QAAQ,CAAC;IAC5C,IAAI,YAAY,IAAI,MAAM;QAAE,OAAO,MAAM,CAAC;IAC1C,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,SAAS,4BAA4B,CAAC,MAAiB;IACrD,MAAM,QAAQ,GAAG,MAAM,CAAC,GAAG,CAAC,QAAQ,EAAE,eAAe,CAAC;IACtD,IAAI,QAAQ,IAAI,IAAI;QAAE,OAAO,QAAQ,CAAC;IACtC,OAAO,yBAAyB,CAAC,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC;AAC3D,CAAC;AAID,SAAS,mBAAmB,CAC1B,eAA+C,EAC/C,YAAoB;IAEpB,IAAI,eAAe,IAAI,IAAI,IAAI,eAAe,KAAK,MAAM,EAAE,CAAC;QAC1D,IAAI,eAAe,KAAK,OAAO;YAAE,OAAO,MAAM,CAAC;QAC/C,IAAI,eAAe,KAAK,SAAS;YAAE,OAAO,SAAS,CAAC;QACpD,OAAO,eAAe,CAAC;IACzB,CAAC;IACD,IAAI,YAAY,IAAI,KAAK;QAAE,OAAO,SAAS,CAAC;IAC5C,IAAI,YAAY,IAAI,KAAK;QAAE,OAAO,KAAK,CAAC;IACxC,IAAI,YAAY,IAAI,MAAM;QAAE,OAAO,QAAQ,CAAC;IAC5C,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,0EAA0E;AAC1E,SAAS,eAAe,CACtB,eAA+C;IAE/C,IAAI,eAAe,IAAI,IAAI,IAAI,eAAe,KAAK,MAAM,EAAE,CAAC;QAC1D,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,IAAI,eAAe,KAAK,SAAS;QAAE,OAAO,KAAK,CAAC;IAChD,IAAI,eAAe,KAAK,OAAO;QAAE,OAAO,OAAO,CAAC;IAChD,OAAO,eAAe,CAAC;AACzB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,oBAAoB,CAClC,MAAiB;IAEjB,IAAI,CAAC,iBAAiB,CAAC,MAAM,CAAC,EAAE,CAAC;QAC/B,IAAI,MAAM,CAAC,GAAG,CAAC,QAAQ,KAAK,WAAW,EAAE,CAAC;YACxC,OAAO,EAAE,SAAS,EAAE,EAAE,sBAAsB,EAAE,IAAI,EAAE,EAAE,CAAC;QACzD,CAAC;QACD,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,MAAM,MAAM,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC;IACtC,MAAM,eAAe,GAAG,MAAM,CAAC,GAAG,CAAC,QAAQ,EAAE,eAAe,CAAC;IAE7D,QAAQ,MAAM,CAAC,GAAG,CAAC,QAAQ,EAAE,CAAC;QAC5B,KAAK,WAAW,CAAC,CAAC,CAAC;YACjB,MAAM,MAAM,GAAG,eAAe,CAAC,eAAe,CAAC,CAAC;YAChD,OAAO;gBACL,SAAS,EAAE;oBACT,sBAAsB,EAAE,IAAI;oBAC5B,QAAQ,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,YAAY,EAAE,MAAM,EAAE;oBACnD,GAAG,CAAC,MAAM,IAAI,EAAE,MAAM,EAAE,CAAC;iBAC1B;aACF,CAAC;QACJ,CAAC;QACD,KAAK,QAAQ;YACX,OAAO;gBACL,MAAM,EAAE;oBACN,eAAe,EAAE,4BAA4B,CAAC,MAAM,CAAC;iBACtD;aACF,CAAC;QACJ,KAAK,WAAW;YACd,OAAO;gBACL,SAAS,EAAE;oBACT,QAAQ,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,YAAY,EAAE,MAAM,EAAE;iBACpD;aACF,CAAC;QACJ,KAAK,QAAQ;YACX,OAAO;gBACL,MAAM,EAAE;oBACN,cAAc,EAAE;wBACd,eAAe,EAAE,IAAI;wBACrB,cAAc,EAAE,MAAM;wBACtB,aAAa,EAAE,mBAAmB,CAAC,eAAe,EAAE,MAAM,CAAC;qBAC5D;iBACF;aACF,CAAC;QACJ,KAAK,QAAQ;YACX,OAAO;gBACL,MAAM,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE;aACxB,CAAC;QACJ,KAAK,YAAY;YACf,OAAO;gBACL,UAAU,EAAE;oBACV,SAAS,EAAE,eAAe;wBACxB,CAAC,CAAC,EAAE,MAAM,EAAE,eAAe,EAAE;wBAC7B,CAAC,CAAC,EAAE,UAAU,EAAE,MAAM,EAAE;iBAC3B;aACF,CAAC;QACJ;YACE,OAAO,SAAS,CAAC;IACrB,CAAC;AACH,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"provider-options.test.d.ts","sourceRoot":"","sources":["../../src/agent/provider-options.test.ts"],"names":[],"mappings":""}
|
|
@@ -0,0 +1,114 @@
|
|
|
1
|
+
import assert from "node:assert/strict";
|
|
2
|
+
import { describe, it } from "node:test";
|
|
3
|
+
import { buildProviderOptions } from "./provider-options.js";
|
|
4
|
+
const base = {
|
|
5
|
+
browser: { headed: false, sessionName: "pqa", defaultTimeout: 25_000, engine: "chrome" },
|
|
6
|
+
skills: { dirs: [], preloads: [] },
|
|
7
|
+
agent: { maxTurns: 10, bashTimeoutMs: 60_000 },
|
|
8
|
+
auth: {},
|
|
9
|
+
};
|
|
10
|
+
function config(provider, thinking) {
|
|
11
|
+
return {
|
|
12
|
+
...base,
|
|
13
|
+
llm: { provider, model: "test-model", thinking },
|
|
14
|
+
};
|
|
15
|
+
}
|
|
16
|
+
describe("buildProviderOptions", () => {
|
|
17
|
+
it("returns undefined when thinking is disabled", () => {
|
|
18
|
+
assert.equal(buildProviderOptions(config("openai", { enabled: false })), undefined);
|
|
19
|
+
});
|
|
20
|
+
it("keeps anthropic parallel tool use off when thinking is disabled", () => {
|
|
21
|
+
assert.deepEqual(buildProviderOptions(config("anthropic", { enabled: false })), { anthropic: { disableParallelToolUse: true } });
|
|
22
|
+
});
|
|
23
|
+
it("enables anthropic extended thinking with budget", () => {
|
|
24
|
+
assert.deepEqual(buildProviderOptions(config("anthropic", { enabled: true, budgetTokens: 8_000 })), {
|
|
25
|
+
anthropic: {
|
|
26
|
+
disableParallelToolUse: true,
|
|
27
|
+
thinking: { type: "enabled", budgetTokens: 8_000 },
|
|
28
|
+
},
|
|
29
|
+
});
|
|
30
|
+
});
|
|
31
|
+
it("maps thinking budget to openai reasoning effort when reasoningEffort is unset", () => {
|
|
32
|
+
assert.deepEqual(buildProviderOptions(config("openai", { enabled: true, budgetTokens: 3_000 })), { openai: { reasoningEffort: "low" } });
|
|
33
|
+
assert.deepEqual(buildProviderOptions(config("openai", { enabled: true, budgetTokens: 25_000 })), { openai: { reasoningEffort: "xhigh" } });
|
|
34
|
+
});
|
|
35
|
+
it("uses explicit reasoningEffort for openai", () => {
|
|
36
|
+
assert.deepEqual(buildProviderOptions(config("openai", {
|
|
37
|
+
enabled: true,
|
|
38
|
+
budgetTokens: 25_000,
|
|
39
|
+
reasoningEffort: "medium",
|
|
40
|
+
})), { openai: { reasoningEffort: "medium" } });
|
|
41
|
+
});
|
|
42
|
+
it("maps reasoningEffort to anthropic effort", () => {
|
|
43
|
+
assert.deepEqual(buildProviderOptions(config("anthropic", {
|
|
44
|
+
enabled: true,
|
|
45
|
+
budgetTokens: 8_000,
|
|
46
|
+
reasoningEffort: "high",
|
|
47
|
+
})), {
|
|
48
|
+
anthropic: {
|
|
49
|
+
disableParallelToolUse: true,
|
|
50
|
+
thinking: { type: "enabled", budgetTokens: 8_000 },
|
|
51
|
+
effort: "high",
|
|
52
|
+
},
|
|
53
|
+
});
|
|
54
|
+
});
|
|
55
|
+
it("omits anthropic effort for none or minimal reasoningEffort", () => {
|
|
56
|
+
assert.deepEqual(buildProviderOptions(config("anthropic", { enabled: true, reasoningEffort: "none" })), {
|
|
57
|
+
anthropic: {
|
|
58
|
+
disableParallelToolUse: true,
|
|
59
|
+
thinking: { type: "enabled", budgetTokens: 10_000 },
|
|
60
|
+
},
|
|
61
|
+
});
|
|
62
|
+
assert.deepEqual(buildProviderOptions(config("anthropic", { enabled: true, reasoningEffort: "minimal" })), {
|
|
63
|
+
anthropic: {
|
|
64
|
+
disableParallelToolUse: true,
|
|
65
|
+
thinking: { type: "enabled", budgetTokens: 10_000 },
|
|
66
|
+
effort: "low",
|
|
67
|
+
},
|
|
68
|
+
});
|
|
69
|
+
});
|
|
70
|
+
it("enables fireworks thinking with budget", () => {
|
|
71
|
+
assert.deepEqual(buildProviderOptions(config("fireworks", { enabled: true, budgetTokens: 12_000 })), {
|
|
72
|
+
fireworks: {
|
|
73
|
+
thinking: { type: "enabled", budgetTokens: 12_000 },
|
|
74
|
+
},
|
|
75
|
+
});
|
|
76
|
+
});
|
|
77
|
+
it("enables google thinking config", () => {
|
|
78
|
+
assert.deepEqual(buildProviderOptions(config("google", {
|
|
79
|
+
enabled: true,
|
|
80
|
+
budgetTokens: 8_000,
|
|
81
|
+
reasoningEffort: "high",
|
|
82
|
+
})), {
|
|
83
|
+
google: {
|
|
84
|
+
thinkingConfig: {
|
|
85
|
+
includeThoughts: true,
|
|
86
|
+
thinkingBudget: 8_000,
|
|
87
|
+
thinkingLevel: "high",
|
|
88
|
+
},
|
|
89
|
+
},
|
|
90
|
+
});
|
|
91
|
+
});
|
|
92
|
+
it("enables ollama think mode", () => {
|
|
93
|
+
assert.deepEqual(buildProviderOptions(config("ollama", { enabled: true })), { ollama: { think: true } });
|
|
94
|
+
});
|
|
95
|
+
it("enables openrouter reasoning with budget", () => {
|
|
96
|
+
assert.deepEqual(buildProviderOptions(config("openrouter", { enabled: true, budgetTokens: 8_000 })), {
|
|
97
|
+
openrouter: {
|
|
98
|
+
reasoning: { max_tokens: 8_000 },
|
|
99
|
+
},
|
|
100
|
+
});
|
|
101
|
+
});
|
|
102
|
+
it("uses explicit reasoningEffort for openrouter", () => {
|
|
103
|
+
assert.deepEqual(buildProviderOptions(config("openrouter", {
|
|
104
|
+
enabled: true,
|
|
105
|
+
budgetTokens: 8_000,
|
|
106
|
+
reasoningEffort: "high",
|
|
107
|
+
})), {
|
|
108
|
+
openrouter: {
|
|
109
|
+
reasoning: { effort: "high" },
|
|
110
|
+
},
|
|
111
|
+
});
|
|
112
|
+
});
|
|
113
|
+
});
|
|
114
|
+
//# sourceMappingURL=provider-options.test.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"provider-options.test.js","sourceRoot":"","sources":["../../src/agent/provider-options.test.ts"],"names":[],"mappings":"AAAA,OAAO,MAAM,MAAM,oBAAoB,CAAC;AACxC,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,WAAW,CAAC;AAEzC,OAAO,EAAE,oBAAoB,EAAE,MAAM,uBAAuB,CAAC;AAE7D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,WAAW,EAAE,KAAK,EAAE,cAAc,EAAE,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE;IACxF,MAAM,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE,QAAQ,EAAE,EAAE,EAAE;IAClC,KAAK,EAAE,EAAE,QAAQ,EAAE,EAAE,EAAE,aAAa,EAAE,MAAM,EAAE;IAC9C,IAAI,EAAE,EAAE;CACoB,CAAC;AAE/B,SAAS,MAAM,CACb,QAAsC,EACtC,QAAuC;IAEvC,OAAO;QACL,GAAG,IAAI;QACP,GAAG,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,YAAY,EAAE,QAAQ,EAAE;KACpC,CAAC;AACjB,CAAC;AAED,QAAQ,CAAC,sBAAsB,EAAE,GAAG,EAAE;IACpC,EAAE,CAAC,6CAA6C,EAAE,GAAG,EAAE;QACrD,MAAM,CAAC,KAAK,CACV,oBAAoB,CAAC,MAAM,CAAC,QAAQ,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC,CAAC,EAC1D,SAAS,CACV,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,iEAAiE,EAAE,GAAG,EAAE;QACzE,MAAM,CAAC,SAAS,CACd,oBAAoB,CAAC,MAAM,CAAC,WAAW,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC,CAAC,EAC7D,EAAE,SAAS,EAAE,EAAE,sBAAsB,EAAE,IAAI,EAAE,EAAE,CAChD,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,iDAAiD,EAAE,GAAG,EAAE;QACzD,MAAM,CAAC,SAAS,CACd,oBAAoB,CAClB,MAAM,CAAC,WAAW,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,KAAK,EAAE,CAAC,CAC5D,EACD;YACE,SAAS,EAAE;gBACT,sBAAsB,EAAE,IAAI;gBAC5B,QAAQ,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,YAAY,EAAE,KAAK,EAAE;aACnD;SACF,CACF,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,+EAA+E,EAAE,GAAG,EAAE;QACvF,MAAM,CAAC,SAAS,CACd,oBAAoB,CAClB,MAAM,CAAC,QAAQ,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,KAAK,EAAE,CAAC,CACzD,EACD,EAAE,MAAM,EAAE,EAAE,eAAe,EAAE,KAAK,EAAE,EAAE,CACvC,CAAC;QACF,MAAM,CAAC,SAAS,CACd,oBAAoB,CAClB,MAAM,CAAC,QAAQ,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,MAAM,EAAE,CAAC,CAC1D,EACD,EAAE,MAAM,EAAE,EAAE,eAAe,EAAE,OAAO,EAAE,EAAE,CACzC,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,0CAA0C,EAAE,GAAG,EAAE;QAClD,MAAM,CAAC,SAAS,CACd,oBAAoB,CAClB,MAAM,CAAC,QAAQ,EAAE;YACf,OAAO,EAAE,IAAI;YACb,YAAY,EAAE,MAAM;YACpB,eAAe,EAAE,QAAQ;SAC1B,CAAC,CACH,EACD,EAAE,MAAM,EAAE,EAAE,eAAe,EAAE,QAAQ,EAAE,EAAE,CAC1C,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,0CAA0C,EAAE,GAAG,EAAE;QAClD,MAAM,CAAC,SAAS,CACd,oBAAoB,CAClB,MAAM,CAAC,WAAW,EAAE;YAClB,OAAO,EAAE,IAAI;YACb,YAAY,EAAE,KAAK;YACnB,eAAe,EAAE,MAAM;SACxB,CAAC,CACH,EACD;YACE,SAAS,EAAE;gBACT,sBAAsB,EAAE,IAAI;gBAC5B,QAAQ,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,YAAY,EAAE,KAAK,EAAE;gBAClD,MAAM,EAAE,MAAM;aACf;SACF,CACF,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4DAA4D,EAAE,GAAG,EAAE;QACpE,MAAM,CAAC,SAAS,CACd,oBAAoB,CAClB,MAAM,CAAC,WAAW,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,eAAe,EAAE,MAAM,EAAE,CAAC,CAChE,EACD;YACE,SAAS,EAAE;gBACT,sBAAsB,EAAE,IAAI;gBAC5B,QAAQ,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,YAAY,EAAE,MAAM,EAAE;aACpD;SACF,CACF,CAAC;QACF,MAAM,CAAC,SAAS,CACd,oBAAoB,CAClB,MAAM,CAAC,WAAW,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,eAAe,EAAE,SAAS,EAAE,CAAC,CACnE,EACD;YACE,SAAS,EAAE;gBACT,sBAAsB,EAAE,IAAI;gBAC5B,QAAQ,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,YAAY,EAAE,MAAM,EAAE;gBACnD,MAAM,EAAE,KAAK;aACd;SACF,CACF,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,wCAAwC,EAAE,GAAG,EAAE;QAChD,MAAM,CAAC,SAAS,CACd,oBAAoB,CAClB,MAAM,CAAC,WAAW,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,MAAM,EAAE,CAAC,CAC7D,EACD;YACE,SAAS,EAAE;gBACT,QAAQ,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,YAAY,EAAE,MAAM,EAAE;aACpD;SACF,CACF,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,gCAAgC,EAAE,GAAG,EAAE;QACxC,MAAM,CAAC,SAAS,CACd,oBAAoB,CAClB,MAAM,CAAC,QAAQ,EAAE;YACf,OAAO,EAAE,IAAI;YACb,YAAY,EAAE,KAAK;YACnB,eAAe,EAAE,MAAM;SACxB,CAAC,CACH,EACD;YACE,MAAM,EAAE;gBACN,cAAc,EAAE;oBACd,eAAe,EAAE,IAAI;oBACrB,cAAc,EAAE,KAAK;oBACrB,aAAa,EAAE,MAAM;iBACtB;aACF;SACF,CACF,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,2BAA2B,EAAE,GAAG,EAAE;QACnC,MAAM,CAAC,SAAS,CACd,oBAAoB,CAAC,MAAM,CAAC,QAAQ,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,EACzD,EAAE,MAAM,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,EAAE,CAC5B,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,0CAA0C,EAAE,GAAG,EAAE;QAClD,MAAM,CAAC,SAAS,CACd,oBAAoB,CAClB,MAAM,CAAC,YAAY,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,KAAK,EAAE,CAAC,CAC7D,EACD;YACE,UAAU,EAAE;gBACV,SAAS,EAAE,EAAE,UAAU,EAAE,KAAK,EAAE;aACjC;SACF,CACF,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8CAA8C,EAAE,GAAG,EAAE;QACtD,MAAM,CAAC,SAAS,CACd,oBAAoB,CAClB,MAAM,CAAC,YAAY,EAAE;YACnB,OAAO,EAAE,IAAI;YACb,YAAY,EAAE,KAAK;YACnB,eAAe,EAAE,MAAM;SACxB,CAAC,CACH,EACD;YACE,UAAU,EAAE;gBACV,SAAS,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE;aAC9B;SACF,CACF,CAAC;IACJ,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}
|
|
@@ -0,0 +1,27 @@
|
|
|
1
|
+
import type { ArtifactsMode, PqaConfig } from "../types/config.js";
|
|
2
|
+
import type { Scenario } from "../types/scenario.js";
|
|
3
|
+
import type { Skill } from "../types/skill.js";
|
|
4
|
+
import type { ScenarioResult, Verdict } from "../types/verdict.js";
|
|
5
|
+
import type { EnvRedactor } from "../redact/env-secrets.js";
|
|
6
|
+
export interface RunScenarioOptions {
|
|
7
|
+
config: PqaConfig;
|
|
8
|
+
skills: Skill[];
|
|
9
|
+
scenario: Scenario;
|
|
10
|
+
cwd: string;
|
|
11
|
+
artifactDir: string;
|
|
12
|
+
authStatePath?: string;
|
|
13
|
+
authProfile?: string;
|
|
14
|
+
profilePath?: string;
|
|
15
|
+
headed: boolean;
|
|
16
|
+
verbose?: boolean;
|
|
17
|
+
artifacts: ArtifactsMode;
|
|
18
|
+
sessionName?: string;
|
|
19
|
+
preparedStartUrl?: string;
|
|
20
|
+
onTurn?: () => Promise<void>;
|
|
21
|
+
redactor?: EnvRedactor;
|
|
22
|
+
noHealing?: boolean;
|
|
23
|
+
scenarioCacheHints?: string;
|
|
24
|
+
}
|
|
25
|
+
export declare function runScenario(options: RunScenarioOptions): Promise<ScenarioResult>;
|
|
26
|
+
export type { Verdict };
|
|
27
|
+
//# sourceMappingURL=runner.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/agent/runner.ts"],"names":[],"mappings":"AAYA,OAAO,KAAK,EAAE,aAAa,EAAE,SAAS,EAAE,MAAM,oBAAoB,CAAC;AACnE,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,sBAAsB,CAAC;AACrD,OAAO,KAAK,EAAE,KAAK,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,KAAK,EAIV,cAAc,EACd,OAAO,EACR,MAAM,qBAAqB,CAAC;AAqB7B,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,0BAA0B,CAAC;AA2B5D,MAAM,WAAW,kBAAkB;IACjC,MAAM,EAAE,SAAS,CAAC;IAClB,MAAM,EAAE,KAAK,EAAE,CAAC;IAChB,QAAQ,EAAE,QAAQ,CAAC;IACnB,GAAG,EAAE,MAAM,CAAC;IACZ,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,MAAM,EAAE,OAAO,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,SAAS,EAAE,aAAa,CAAC;IACzB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,MAAM,CAAC,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAC;IAC7B,QAAQ,CAAC,EAAE,WAAW,CAAC;IACvB,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,kBAAkB,CAAC,EAAE,MAAM,CAAC;CAC7B;AA+DD,wBAAsB,WAAW,CAC/B,OAAO,EAAE,kBAAkB,GAC1B,OAAO,CAAC,cAAc,CAAC,CA2RzB;AAED,YAAY,EAAE,OAAO,EAAE,CAAC"}
|
|
@@ -0,0 +1,291 @@
|
|
|
1
|
+
import { generateText, stepCountIs, tool, } from "ai";
|
|
2
|
+
import { z } from "zod";
|
|
3
|
+
import { resolveHealingConfig } from "../config/load.js";
|
|
4
|
+
import { classifyFailure, isHealingEnabled, isRecoveryAllowed, } from "../healing/classify.js";
|
|
5
|
+
import { buildRecoveryPrompt } from "../healing/recovery-prompt.js";
|
|
6
|
+
import { buildBrowserEnv, runBash } from "./bash.js";
|
|
7
|
+
import { buildInitialPrompt, buildSystemPrompt } from "./prompt.js";
|
|
8
|
+
import { buildVerdictRetryPrompt } from "./verdict-retry-prompt.js";
|
|
9
|
+
import { writeTranscript } from "../reporter/index.js";
|
|
10
|
+
import { appendFinalTextToTranscript, appendStepToTranscript, appendTranscriptMessage, extractVerdict, formatStepForTranscript, stripLastAssistantTurn, } from "./verdict.js";
|
|
11
|
+
import { resolveStatePath } from "../auth/store.js";
|
|
12
|
+
import { createLlmModel } from "./llm-model.js";
|
|
13
|
+
import { buildProviderOptions } from "./provider-options.js";
|
|
14
|
+
const MAX_VERDICT_RETRIES = 5;
|
|
15
|
+
/** Extra steps when re-emitting a verdict after an invalid completion. */
|
|
16
|
+
const VERDICT_RETRY_MAX_STEPS = 10;
|
|
17
|
+
function removeLastAssistantMessage(transcript) {
|
|
18
|
+
while (transcript.entries.at(-1)?.type === "bash") {
|
|
19
|
+
transcript.entries.pop();
|
|
20
|
+
}
|
|
21
|
+
const last = transcript.entries.at(-1);
|
|
22
|
+
if (last?.type === "message" && last.role === "assistant") {
|
|
23
|
+
transcript.entries.pop();
|
|
24
|
+
}
|
|
25
|
+
}
|
|
26
|
+
function persistTranscript(options, transcript) {
|
|
27
|
+
if (options.verbose) {
|
|
28
|
+
writeTranscript(options.artifactDir, transcript, options.redactor);
|
|
29
|
+
}
|
|
30
|
+
}
|
|
31
|
+
async function retryVerdictCompletion(options) {
|
|
32
|
+
let { result, finalText } = options;
|
|
33
|
+
for (let attempt = 0; !extractVerdict(finalText) && attempt < MAX_VERDICT_RETRIES; attempt++) {
|
|
34
|
+
removeLastAssistantMessage(options.transcript);
|
|
35
|
+
const retryPrompt = buildVerdictRetryPrompt(options.runOptions.scenario);
|
|
36
|
+
appendTranscriptMessage(options.transcript, "user", retryPrompt);
|
|
37
|
+
persistTranscript(options.runOptions, options.transcript);
|
|
38
|
+
const retryMessages = [
|
|
39
|
+
...stripLastAssistantTurn(result.response.messages),
|
|
40
|
+
{ role: "user", content: retryPrompt },
|
|
41
|
+
];
|
|
42
|
+
options.stepTiming.startMs = Date.now();
|
|
43
|
+
result = await generateText({
|
|
44
|
+
model: createLlmModel(options.config),
|
|
45
|
+
system: options.system,
|
|
46
|
+
messages: retryMessages,
|
|
47
|
+
providerOptions: options.providerOptions,
|
|
48
|
+
stopWhen: stepCountIs(VERDICT_RETRY_MAX_STEPS),
|
|
49
|
+
onStepFinish: options.onStepFinish,
|
|
50
|
+
});
|
|
51
|
+
finalText = result.text || finalText;
|
|
52
|
+
appendFinalTextToTranscript(options.transcript, options.runOptions.redactor
|
|
53
|
+
? options.runOptions.redactor.redact(finalText)
|
|
54
|
+
: finalText, { durationMs: Date.now() - options.stepTiming.startMs });
|
|
55
|
+
options.stepTiming.startMs = Date.now();
|
|
56
|
+
persistTranscript(options.runOptions, options.transcript);
|
|
57
|
+
}
|
|
58
|
+
return { result, finalText };
|
|
59
|
+
}
|
|
60
|
+
export async function runScenario(options) {
|
|
61
|
+
const start = Date.now();
|
|
62
|
+
const transcript = { entries: [] };
|
|
63
|
+
const sessionName = options.sessionName ?? options.config.browser.sessionName;
|
|
64
|
+
const authSavePath = options.authProfile
|
|
65
|
+
? resolveStatePath(options.cwd, options.authProfile, options.config)
|
|
66
|
+
: undefined;
|
|
67
|
+
const bashEnv = buildBrowserEnv({
|
|
68
|
+
cwd: options.cwd,
|
|
69
|
+
headed: options.headed,
|
|
70
|
+
sessionName,
|
|
71
|
+
engine: options.config.browser.engine,
|
|
72
|
+
lightpanda: options.config.browser.lightpanda,
|
|
73
|
+
profilePath: options.profilePath,
|
|
74
|
+
authStatePath: options.profilePath ? undefined : options.authStatePath,
|
|
75
|
+
authSavePath,
|
|
76
|
+
artifactDir: options.artifactDir,
|
|
77
|
+
});
|
|
78
|
+
const system = buildSystemPrompt(options.config, options.skills, options.scenario, {
|
|
79
|
+
cwd: options.cwd,
|
|
80
|
+
artifactDir: options.artifactDir,
|
|
81
|
+
authStatePath: options.authStatePath,
|
|
82
|
+
authProfile: options.authProfile,
|
|
83
|
+
profilePath: options.profilePath,
|
|
84
|
+
headed: options.headed,
|
|
85
|
+
sessionName,
|
|
86
|
+
artifacts: options.artifacts,
|
|
87
|
+
scenarioCacheHints: options.scenarioCacheHints,
|
|
88
|
+
preparedStartUrl: options.preparedStartUrl,
|
|
89
|
+
});
|
|
90
|
+
let finalText = "";
|
|
91
|
+
let turn = 0;
|
|
92
|
+
const stepTiming = { startMs: Date.now() };
|
|
93
|
+
const pendingBashEntries = [];
|
|
94
|
+
const tools = {
|
|
95
|
+
bash: tool({
|
|
96
|
+
description: "Run ONE bash command. For UI interactions (click, fill, select, open, press), run a single agent-browser command per call. Use snapshot -i before acting on refs.",
|
|
97
|
+
inputSchema: z.object({
|
|
98
|
+
command: z.string().describe("Shell command to execute"),
|
|
99
|
+
}),
|
|
100
|
+
execute: async ({ command }) => {
|
|
101
|
+
const entry = await runBash(command, {
|
|
102
|
+
cwd: options.cwd,
|
|
103
|
+
timeoutMs: options.config.agent.bashTimeoutMs,
|
|
104
|
+
env: bashEnv,
|
|
105
|
+
});
|
|
106
|
+
const redacted = options.redactor
|
|
107
|
+
? options.redactor.redactBashEntry(entry)
|
|
108
|
+
: entry;
|
|
109
|
+
pendingBashEntries.push(redacted);
|
|
110
|
+
if (options.verbose) {
|
|
111
|
+
console.log(`\n$ ${redacted.command}`);
|
|
112
|
+
if (redacted.stdout)
|
|
113
|
+
console.log(redacted.stdout.slice(0, 2000));
|
|
114
|
+
if (redacted.stderr)
|
|
115
|
+
console.error(redacted.stderr.slice(0, 1000));
|
|
116
|
+
}
|
|
117
|
+
return {
|
|
118
|
+
exitCode: entry.exitCode,
|
|
119
|
+
stdout: entry.stdout.slice(0, 8000),
|
|
120
|
+
stderr: entry.stderr.slice(0, 2000),
|
|
121
|
+
};
|
|
122
|
+
},
|
|
123
|
+
}),
|
|
124
|
+
};
|
|
125
|
+
const initialPrompt = buildInitialPrompt(options.scenario, options.preparedStartUrl);
|
|
126
|
+
appendTranscriptMessage(transcript, "user", initialPrompt);
|
|
127
|
+
const onStepFinish = async (step) => {
|
|
128
|
+
turn += 1;
|
|
129
|
+
const recordedAt = new Date();
|
|
130
|
+
const durationMs = recordedAt.getTime() - stepTiming.startMs;
|
|
131
|
+
stepTiming.startMs = recordedAt.getTime();
|
|
132
|
+
const bashEntries = pendingBashEntries.splice(0);
|
|
133
|
+
const stepInput = {
|
|
134
|
+
text: step.text,
|
|
135
|
+
reasoningText: step.reasoningText,
|
|
136
|
+
toolCalls: step.toolCalls,
|
|
137
|
+
};
|
|
138
|
+
const formatted = formatStepForTranscript(stepInput);
|
|
139
|
+
const safeFormatted = options.redactor
|
|
140
|
+
? {
|
|
141
|
+
content: formatted.content
|
|
142
|
+
? options.redactor.redact(formatted.content)
|
|
143
|
+
: formatted.content,
|
|
144
|
+
thinking: formatted.thinking
|
|
145
|
+
? options.redactor.redact(formatted.thinking)
|
|
146
|
+
: formatted.thinking,
|
|
147
|
+
}
|
|
148
|
+
: formatted;
|
|
149
|
+
const changed = appendStepToTranscript(transcript, stepInput, bashEntries, safeFormatted, { at: recordedAt, durationMs });
|
|
150
|
+
if (changed) {
|
|
151
|
+
persistTranscript(options, transcript);
|
|
152
|
+
}
|
|
153
|
+
if (step.text)
|
|
154
|
+
finalText = step.text;
|
|
155
|
+
if (options.onTurn)
|
|
156
|
+
await options.onTurn();
|
|
157
|
+
};
|
|
158
|
+
try {
|
|
159
|
+
const providerOptions = buildProviderOptions(options.config);
|
|
160
|
+
stepTiming.startMs = Date.now();
|
|
161
|
+
let result = (await generateText({
|
|
162
|
+
model: createLlmModel(options.config),
|
|
163
|
+
system,
|
|
164
|
+
prompt: initialPrompt,
|
|
165
|
+
tools,
|
|
166
|
+
providerOptions,
|
|
167
|
+
stopWhen: stepCountIs(options.config.agent.maxTurns),
|
|
168
|
+
onStepFinish,
|
|
169
|
+
}));
|
|
170
|
+
finalText = result.text || finalText;
|
|
171
|
+
appendFinalTextToTranscript(transcript, options.redactor ? options.redactor.redact(finalText) : finalText, { durationMs: Date.now() - stepTiming.startMs });
|
|
172
|
+
stepTiming.startMs = Date.now();
|
|
173
|
+
persistTranscript(options, transcript);
|
|
174
|
+
({ result, finalText } = await retryVerdictCompletion({
|
|
175
|
+
config: options.config,
|
|
176
|
+
system,
|
|
177
|
+
providerOptions,
|
|
178
|
+
result,
|
|
179
|
+
transcript,
|
|
180
|
+
runOptions: options,
|
|
181
|
+
onStepFinish,
|
|
182
|
+
finalText,
|
|
183
|
+
stepTiming,
|
|
184
|
+
}));
|
|
185
|
+
let verdict = extractVerdict(finalText);
|
|
186
|
+
const healingMeta = {
|
|
187
|
+
used: false,
|
|
188
|
+
recoveryTurns: 0,
|
|
189
|
+
scenarioRetries: 0,
|
|
190
|
+
};
|
|
191
|
+
let draftResult = {
|
|
192
|
+
scenario: options.scenario.frontmatter.name,
|
|
193
|
+
filePath: options.scenario.filePath,
|
|
194
|
+
status: verdict?.status === "pass" ? "pass" : "fail",
|
|
195
|
+
durationMs: Date.now() - start,
|
|
196
|
+
verdict: options.redactor ? options.redactor.redactVerdict(verdict) : verdict,
|
|
197
|
+
transcript,
|
|
198
|
+
artifactDir: options.artifactDir,
|
|
199
|
+
error: verdict
|
|
200
|
+
? undefined
|
|
201
|
+
: "Agent did not emit a valid verdict JSON block",
|
|
202
|
+
healing: healingMeta,
|
|
203
|
+
};
|
|
204
|
+
if (isHealingEnabled(options.config, options.noHealing) &&
|
|
205
|
+
verdict?.status === "fail") {
|
|
206
|
+
const classified = classifyFailure(draftResult, options.scenario, options.config);
|
|
207
|
+
healingMeta.failureKind = classified.kind;
|
|
208
|
+
healingMeta.signals = classified.signals;
|
|
209
|
+
if (isRecoveryAllowed(classified, options.config, options.noHealing)) {
|
|
210
|
+
const maxRecovery = resolveHealingConfig(options.config).maxRecoveryTurns;
|
|
211
|
+
const failed = verdict.checkpoints.filter((c) => !c.pass);
|
|
212
|
+
for (let recoveryAttempt = 0; recoveryAttempt < maxRecovery; recoveryAttempt++) {
|
|
213
|
+
const recoveryPrompt = buildRecoveryPrompt(failed);
|
|
214
|
+
appendTranscriptMessage(transcript, "user", recoveryPrompt);
|
|
215
|
+
persistTranscript(options, transcript);
|
|
216
|
+
const recoveryMessages = [
|
|
217
|
+
...result.response.messages,
|
|
218
|
+
{ role: "user", content: recoveryPrompt },
|
|
219
|
+
];
|
|
220
|
+
stepTiming.startMs = Date.now();
|
|
221
|
+
result = (await generateText({
|
|
222
|
+
model: createLlmModel(options.config),
|
|
223
|
+
system,
|
|
224
|
+
messages: recoveryMessages,
|
|
225
|
+
tools,
|
|
226
|
+
providerOptions,
|
|
227
|
+
stopWhen: stepCountIs(options.config.agent.maxTurns),
|
|
228
|
+
onStepFinish,
|
|
229
|
+
}));
|
|
230
|
+
finalText = result.text || finalText;
|
|
231
|
+
appendFinalTextToTranscript(transcript, options.redactor ? options.redactor.redact(finalText) : finalText, { durationMs: Date.now() - stepTiming.startMs });
|
|
232
|
+
stepTiming.startMs = Date.now();
|
|
233
|
+
persistTranscript(options, transcript);
|
|
234
|
+
({ result, finalText } = await retryVerdictCompletion({
|
|
235
|
+
config: options.config,
|
|
236
|
+
system,
|
|
237
|
+
providerOptions,
|
|
238
|
+
result,
|
|
239
|
+
transcript,
|
|
240
|
+
runOptions: options,
|
|
241
|
+
onStepFinish,
|
|
242
|
+
finalText,
|
|
243
|
+
stepTiming,
|
|
244
|
+
}));
|
|
245
|
+
verdict = extractVerdict(finalText);
|
|
246
|
+
healingMeta.recoveryTurns += 1;
|
|
247
|
+
draftResult = {
|
|
248
|
+
...draftResult,
|
|
249
|
+
status: verdict?.status === "pass" ? "pass" : "fail",
|
|
250
|
+
verdict: options.redactor
|
|
251
|
+
? options.redactor.redactVerdict(verdict)
|
|
252
|
+
: verdict,
|
|
253
|
+
error: verdict
|
|
254
|
+
? undefined
|
|
255
|
+
: "Agent did not emit a valid verdict JSON block",
|
|
256
|
+
};
|
|
257
|
+
if (verdict?.status === "pass") {
|
|
258
|
+
healingMeta.used = true;
|
|
259
|
+
break;
|
|
260
|
+
}
|
|
261
|
+
}
|
|
262
|
+
}
|
|
263
|
+
}
|
|
264
|
+
if (!draftResult.verdict) {
|
|
265
|
+
draftResult.status = "fail";
|
|
266
|
+
}
|
|
267
|
+
return {
|
|
268
|
+
...draftResult,
|
|
269
|
+
durationMs: Date.now() - start,
|
|
270
|
+
healing: healingMeta,
|
|
271
|
+
};
|
|
272
|
+
}
|
|
273
|
+
catch (err) {
|
|
274
|
+
if (pendingBashEntries.length > 0) {
|
|
275
|
+
appendStepToTranscript(transcript, { text: "", toolCalls: [] }, pendingBashEntries.splice(0));
|
|
276
|
+
}
|
|
277
|
+
persistTranscript(options, transcript);
|
|
278
|
+
const error = String(err);
|
|
279
|
+
return {
|
|
280
|
+
scenario: options.scenario.frontmatter.name,
|
|
281
|
+
filePath: options.scenario.filePath,
|
|
282
|
+
status: "error",
|
|
283
|
+
durationMs: Date.now() - start,
|
|
284
|
+
verdict: null,
|
|
285
|
+
transcript,
|
|
286
|
+
artifactDir: options.artifactDir,
|
|
287
|
+
error: options.redactor ? options.redactor.redact(error) : error,
|
|
288
|
+
};
|
|
289
|
+
}
|
|
290
|
+
}
|
|
291
|
+
//# sourceMappingURL=runner.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"runner.js","sourceRoot":"","sources":["../../src/agent/runner.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,YAAY,EAEZ,WAAW,EACX,IAAI,GAIL,MAAM,IAAI,CAAC;AAGZ,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAWxB,OAAO,EAAE,oBAAoB,EAAE,MAAM,mBAAmB,CAAC;AACzD,OAAO,EACL,eAAe,EACf,gBAAgB,EAChB,iBAAiB,GAClB,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAAE,mBAAmB,EAAE,MAAM,+BAA+B,CAAC;AACpE,OAAO,EAAE,eAAe,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACrD,OAAO,EAAE,kBAAkB,EAAE,iBAAiB,EAAE,MAAM,aAAa,CAAC;AACpE,OAAO,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AACpE,OAAO,EAAE,eAAe,EAAE,MAAM,sBAAsB,CAAC;AACvD,OAAO,EACL,2BAA2B,EAC3B,sBAAsB,EACtB,uBAAuB,EACvB,cAAc,EACd,uBAAuB,EACvB,sBAAsB,GACvB,MAAM,cAAc,CAAC;AACtB,OAAO,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AAEpD,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAChD,OAAO,EAAE,oBAAoB,EAAE,MAAM,uBAAuB,CAAC;AAE7D,MAAM,mBAAmB,GAAG,CAAC,CAAC;AAC9B,0EAA0E;AAC1E,MAAM,uBAAuB,GAAG,EAAE,CAAC;AAEnC,SAAS,0BAA0B,CAAC,UAA2B;IAC7D,OAAO,UAAU,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,IAAI,KAAK,MAAM,EAAE,CAAC;QAClD,UAAU,CAAC,OAAO,CAAC,GAAG,EAAE,CAAC;IAC3B,CAAC;IACD,MAAM,IAAI,GAAG,UAAU,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IACvC,IAAI,IAAI,EAAE,IAAI,KAAK,SAAS,IAAI,IAAI,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;QAC1D,UAAU,CAAC,OAAO,CAAC,GAAG,EAAE,CAAC;IAC3B,CAAC;AACH,CAAC;AAED,SAAS,iBAAiB,CACxB,OAA2B,EAC3B,UAA2B;IAE3B,IAAI,OAAO,CAAC,OAAO,EAAE,CAAC;QACpB,eAAe,CAAC,OAAO,CAAC,WAAW,EAAE,UAAU,EAAE,OAAO,CAAC,QAAQ,CAAC,CAAC;IACrE,CAAC;AACH,CAAC;AAsBD,KAAK,UAAU,sBAAsB,CAAC,OAcrC;IAIC,IAAI,EAAE,MAAM,EAAE,SAAS,EAAE,GAAG,OAAO,CAAC;IAEpC,KACE,IAAI,OAAO,GAAG,CAAC,EACf,CAAC,cAAc,CAAC,SAAS,CAAC,IAAI,OAAO,GAAG,mBAAmB,EAC3D,OAAO,EAAE,EACT,CAAC;QACD,0BAA0B,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC;QAE/C,MAAM,WAAW,GAAG,uBAAuB,CAAC,OAAO,CAAC,UAAU,CAAC,QAAQ,CAAC,CAAC;QACzE,uBAAuB,CAAC,OAAO,CAAC,UAAU,EAAE,MAAM,EAAE,WAAW,CAAC,CAAC;QACjE,iBAAiB,CAAC,OAAO,CAAC,UAAU,EAAE,OAAO,CAAC,UAAU,CAAC,CAAC;QAE1D,MAAM,aAAa,GAAmB;YACpC,GAAG,sBAAsB,CAAC,MAAM,CAAC,QAAQ,CAAC,QAA0B,CAAC;YACrE,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,WAAW,EAAE;SACvC,CAAC;QAEF,OAAO,CAAC,UAAU,CAAC,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACxC,MAAM,GAAG,MAAM,YAAY,CAAC;YAC1B,KAAK,EAAE,cAAc,CAAC,OAAO,CAAC,MAAM,CAAC;YACrC,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,QAAQ,EAAE,aAAa;YACvB,eAAe,EAAE,OAAO,CAAC,eAAe;YACxC,QAAQ,EAAE,WAAW,CAAC,uBAAuB,CAAC;YAC9C,YAAY,EAAE,OAAO,CAAC,YAAY;SACnC,CAAC,CAAC;QAEH,SAAS,GAAG,MAAM,CAAC,IAAI,IAAI,SAAS,CAAC;QACrC,2BAA2B,CACzB,OAAO,CAAC,UAAU,EAClB,OAAO,CAAC,UAAU,CAAC,QAAQ;YACzB,CAAC,CAAC,OAAO,CAAC,UAAU,CAAC,QAAQ,CAAC,MAAM,CAAC,SAAS,CAAC;YAC/C,CAAC,CAAC,SAAS,EACb,EAAE,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,OAAO,CAAC,UAAU,CAAC,OAAO,EAAE,CACxD,CAAC;QACF,OAAO,CAAC,UAAU,CAAC,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACxC,iBAAiB,CAAC,OAAO,CAAC,UAAU,EAAE,OAAO,CAAC,UAAU,CAAC,CAAC;IAC5D,CAAC;IAED,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC;AAC/B,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,WAAW,CAC/B,OAA2B;IAE3B,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IACzB,MAAM,UAAU,GAAoB,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC;IACpD,MAAM,WAAW,GACf,OAAO,CAAC,WAAW,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO,CAAC,WAAW,CAAC;IAC5D,MAAM,YAAY,GAAG,OAAO,CAAC,WAAW;QACtC,CAAC,CAAC,gBAAgB,CAAC,OAAO,CAAC,GAAG,EAAE,OAAO,CAAC,WAAW,EAAE,OAAO,CAAC,MAAM,CAAC;QACpE,CAAC,CAAC,SAAS,CAAC;IACd,MAAM,OAAO,GAAG,eAAe,CAAC;QAC9B,GAAG,EAAE,OAAO,CAAC,GAAG;QAChB,MAAM,EAAE,OAAO,CAAC,MAAM;QACtB,WAAW;QACX,MAAM,EAAE,OAAO,CAAC,MAAM,CAAC,OAAO,CAAC,MAAM;QACrC,UAAU,EAAE,OAAO,CAAC,MAAM,CAAC,OAAO,CAAC,UAAU;QAC7C,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,aAAa,EAAE,OAAO,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,aAAa;QACtE,YAAY;QACZ,WAAW,EAAE,OAAO,CAAC,WAAW;KACjC,CAAC,CAAC;IAEH,MAAM,MAAM,GAAG,iBAAiB,CAC9B,OAAO,CAAC,MAAM,EACd,OAAO,CAAC,MAAM,EACd,OAAO,CAAC,QAAQ,EAChB;QACE,GAAG,EAAE,OAAO,CAAC,GAAG;QAChB,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,aAAa,EAAE,OAAO,CAAC,aAAa;QACpC,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,MAAM,EAAE,OAAO,CAAC,MAAM;QACtB,WAAW;QACX,SAAS,EAAE,OAAO,CAAC,SAAS;QAC5B,kBAAkB,EAAE,OAAO,CAAC,kBAAkB;QAC9C,gBAAgB,EAAE,OAAO,CAAC,gBAAgB;KAC3C,CACF,CAAC;IAEF,IAAI,SAAS,GAAG,EAAE,CAAC;IACnB,IAAI,IAAI,GAAG,CAAC,CAAC;IACb,MAAM,UAAU,GAAG,EAAE,OAAO,EAAE,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC;IAC3C,MAAM,kBAAkB,GAAgB,EAAE,CAAC;IAE3C,MAAM,KAAK,GAAG;QACZ,IAAI,EAAE,IAAI,CAAC;YACT,WAAW,EACT,mKAAmK;YACrK,WAAW,EAAE,CAAC,CAAC,MAAM,CAAC;gBACpB,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,0BAA0B,CAAC;aACzD,CAAC;YACF,OAAO,EAAE,KAAK,EAAE,EAAE,OAAO,EAAE,EAAE,EAAE;gBAC7B,MAAM,KAAK,GAAG,MAAM,OAAO,CAAC,OAAO,EAAE;oBACnC,GAAG,EAAE,OAAO,CAAC,GAAG;oBAChB,SAAS,EAAE,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,aAAa;oBAC7C,GAAG,EAAE,OAAO;iBACb,CAAC,CAAC;gBACH,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ;oBAC/B,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAC,KAAK,CAAC;oBACzC,CAAC,CAAC,KAAK,CAAC;gBACV,kBAAkB,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;gBAClC,IAAI,OAAO,CAAC,OAAO,EAAE,CAAC;oBACpB,OAAO,CAAC,GAAG,CAAC,OAAO,QAAQ,CAAC,OAAO,EAAE,CAAC,CAAC;oBACvC,IAAI,QAAQ,CAAC,MAAM;wBAAE,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC;oBACjE,IAAI,QAAQ,CAAC,MAAM;wBAAE,OAAO,CAAC,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC;gBACrE,CAAC;gBACD,OAAO;oBACL,QAAQ,EAAE,KAAK,CAAC,QAAQ;oBACxB,MAAM,EAAE,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC;oBACnC,MAAM,EAAE,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC;iBACpC,CAAC;YACJ,CAAC;SACF,CAAC;KACH,CAAC;IAEF,MAAM,aAAa,GAAG,kBAAkB,CACtC,OAAO,CAAC,QAAQ,EAChB,OAAO,CAAC,gBAAgB,CACzB,CAAC;IACF,uBAAuB,CAAC,UAAU,EAAE,MAAM,EAAE,aAAa,CAAC,CAAC;IAE3D,MAAM,YAAY,GAAG,KAAK,EAAE,IAI3B,EAAE,EAAE;QACH,IAAI,IAAI,CAAC,CAAC;QACV,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC;QAC9B,MAAM,UAAU,GAAG,UAAU,CAAC,OAAO,EAAE,GAAG,UAAU,CAAC,OAAO,CAAC;QAC7D,UAAU,CAAC,OAAO,GAAG,UAAU,CAAC,OAAO,EAAE,CAAC;QAC1C,MAAM,WAAW,GAAG,kBAAkB,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;QACjD,MAAM,SAAS,GAAG;YAChB,IAAI,EAAE,IAAI,CAAC,IAAI;YACf,aAAa,EAAE,IAAI,CAAC,aAAa;YACjC,SAAS,EAAE,IAAI,CAAC,SAAS;SAC1B,CAAC;QACF,MAAM,SAAS,GAAG,uBAAuB,CAAC,SAAS,CAAC,CAAC;QACrD,MAAM,aAAa,GAAG,OAAO,CAAC,QAAQ;YACpC,CAAC,CAAC;gBACA,OAAO,EAAE,SAAS,CAAC,OAAO;oBACxB,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,SAAS,CAAC,OAAO,CAAC;oBAC5C,CAAC,CAAC,SAAS,CAAC,OAAO;gBACrB,QAAQ,EAAE,SAAS,CAAC,QAAQ;oBAC1B,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,SAAS,CAAC,QAAQ,CAAC;oBAC7C,CAAC,CAAC,SAAS,CAAC,QAAQ;aACvB;YACD,CAAC,CAAC,SAAS,CAAC;QACd,MAAM,OAAO,GAAG,sBAAsB,CACpC,UAAU,EACV,SAAS,EACT,WAAW,EACX,aAAa,EACb,EAAE,EAAE,EAAE,UAAU,EAAE,UAAU,EAAE,CAC/B,CAAC;QACF,IAAI,OAAO,EAAE,CAAC;YACZ,iBAAiB,CAAC,OAAO,EAAE,UAAU,CAAC,CAAC;QACzC,CAAC;QACD,IAAI,IAAI,CAAC,IAAI;YAAE,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC;QACrC,IAAI,OAAO,CAAC,MAAM;YAAE,MAAM,OAAO,CAAC,MAAM,EAAE,CAAC;IAC7C,CAAC,CAAC;IAEF,IAAI,CAAC;QACH,MAAM,eAAe,GAAG,oBAAoB,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QAE7D,UAAU,CAAC,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAChC,IAAI,MAAM,GAAG,CAAC,MAAM,YAAY,CAAC;YAC/B,KAAK,EAAE,cAAc,CAAC,OAAO,CAAC,MAAM,CAAC;YACrC,MAAM;YACN,MAAM,EAAE,aAAa;YACrB,KAAK;YACL,eAAe;YACf,QAAQ,EAAE,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,QAAQ,CAAC;YACpD,YAAY;SACb,CAAC,CAA+D,CAAC;QAElE,SAAS,GAAG,MAAM,CAAC,IAAI,IAAI,SAAS,CAAC;QACrC,2BAA2B,CACzB,UAAU,EACV,OAAO,CAAC,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,EACjE,EAAE,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,UAAU,CAAC,OAAO,EAAE,CAChD,CAAC;QACF,UAAU,CAAC,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAChC,iBAAiB,CAAC,OAAO,EAAE,UAAU,CAAC,CAAC;QAEvC,CAAC,EAAE,MAAM,EAAE,SAAS,EAAE,GAAG,MAAM,sBAAsB,CAAC;YACpD,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,MAAM;YACN,eAAe;YACf,MAAM;YACN,UAAU;YACV,UAAU,EAAE,OAAO;YACnB,YAAY;YACZ,SAAS;YACT,UAAU;SACX,CAAC,CAAC,CAAC;QAEJ,IAAI,OAAO,GAAG,cAAc,CAAC,SAAS,CAAC,CAAC;QAExC,MAAM,WAAW,GAAgB;YAC/B,IAAI,EAAE,KAAK;YACX,aAAa,EAAE,CAAC;YAChB,eAAe,EAAE,CAAC;SACnB,CAAC;QAEF,IAAI,WAAW,GAAmB;YAChC,QAAQ,EAAE,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAC,IAAI;YAC3C,QAAQ,EAAE,OAAO,CAAC,QAAQ,CAAC,QAAQ;YACnC,MAAM,EAAE,OAAO,EAAE,MAAM,KAAK,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM;YACpD,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;YAC9B,OAAO,EAAE,OAAO,CAAC,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO;YAC7E,UAAU;YACV,WAAW,EAAE,OAAO,CAAC,WAAW;YAChC,KAAK,EAAE,OAAO;gBACZ,CAAC,CAAC,SAAS;gBACX,CAAC,CAAC,+CAA+C;YACnD,OAAO,EAAE,WAAW;SACrB,CAAC;QAEF,IACE,gBAAgB,CAAC,OAAO,CAAC,MAAM,EAAE,OAAO,CAAC,SAAS,CAAC;YACnD,OAAO,EAAE,MAAM,KAAK,MAAM,EAC1B,CAAC;YACD,MAAM,UAAU,GAAG,eAAe,CAChC,WAAW,EACX,OAAO,CAAC,QAAQ,EAChB,OAAO,CAAC,MAAM,CACf,CAAC;YACF,WAAW,CAAC,WAAW,GAAG,UAAU,CAAC,IAAI,CAAC;YAC1C,WAAW,CAAC,OAAO,GAAG,UAAU,CAAC,OAAO,CAAC;YAEzC,IAAI,iBAAiB,CAAC,UAAU,EAAE,OAAO,CAAC,MAAM,EAAE,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC;gBACrE,MAAM,WAAW,GAAG,oBAAoB,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,gBAAgB,CAAC;gBAC1E,MAAM,MAAM,GAAG,OAAO,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;gBAE1D,KAAK,IAAI,eAAe,GAAG,CAAC,EAAE,eAAe,GAAG,WAAW,EAAE,eAAe,EAAE,EAAE,CAAC;oBAC/E,MAAM,cAAc,GAAG,mBAAmB,CAAC,MAAM,CAAC,CAAC;oBACnD,uBAAuB,CAAC,UAAU,EAAE,MAAM,EAAE,cAAc,CAAC,CAAC;oBAC5D,iBAAiB,CAAC,OAAO,EAAE,UAAU,CAAC,CAAC;oBAEvC,MAAM,gBAAgB,GAAmB;wBACvC,GAAG,MAAM,CAAC,QAAQ,CAAC,QAAQ;wBAC3B,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,cAAc,EAAE;qBAC1C,CAAC;oBAEF,UAAU,CAAC,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;oBAChC,MAAM,GAAG,CAAC,MAAM,YAAY,CAAC;wBAC3B,KAAK,EAAE,cAAc,CAAC,OAAO,CAAC,MAAM,CAAC;wBACrC,MAAM;wBACN,QAAQ,EAAE,gBAAgB;wBAC1B,KAAK;wBACL,eAAe;wBACf,QAAQ,EAAE,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,QAAQ,CAAC;wBACpD,YAAY;qBACb,CAAC,CAA+D,CAAC;oBAElE,SAAS,GAAG,MAAM,CAAC,IAAI,IAAI,SAAS,CAAC;oBACrC,2BAA2B,CACzB,UAAU,EACV,OAAO,CAAC,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,EACjE,EAAE,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,UAAU,CAAC,OAAO,EAAE,CAChD,CAAC;oBACF,UAAU,CAAC,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;oBAChC,iBAAiB,CAAC,OAAO,EAAE,UAAU,CAAC,CAAC;oBAEvC,CAAC,EAAE,MAAM,EAAE,SAAS,EAAE,GAAG,MAAM,sBAAsB,CAAC;wBACpD,MAAM,EAAE,OAAO,CAAC,MAAM;wBACtB,MAAM;wBACN,eAAe;wBACf,MAAM;wBACN,UAAU;wBACV,UAAU,EAAE,OAAO;wBACnB,YAAY;wBACZ,SAAS;wBACT,UAAU;qBACX,CAAC,CAAC,CAAC;oBACJ,OAAO,GAAG,cAAc,CAAC,SAAS,CAAC,CAAC;oBAEpC,WAAW,CAAC,aAAa,IAAI,CAAC,CAAC;oBAE/B,WAAW,GAAG;wBACZ,GAAG,WAAW;wBACd,MAAM,EAAE,OAAO,EAAE,MAAM,KAAK,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM;wBACpD,OAAO,EAAE,OAAO,CAAC,QAAQ;4BACvB,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAC,OAAO,CAAC;4BACzC,CAAC,CAAC,OAAO;wBACX,KAAK,EAAE,OAAO;4BACZ,CAAC,CAAC,SAAS;4BACX,CAAC,CAAC,+CAA+C;qBACpD,CAAC;oBAEF,IAAI,OAAO,EAAE,MAAM,KAAK,MAAM,EAAE,CAAC;wBAC/B,WAAW,CAAC,IAAI,GAAG,IAAI,CAAC;wBACxB,MAAM;oBACR,CAAC;gBACH,CAAC;YACH,CAAC;QACH,CAAC;QAED,IAAI,CAAC,WAAW,CAAC,OAAO,EAAE,CAAC;YACzB,WAAW,CAAC,MAAM,GAAG,MAAM,CAAC;QAC9B,CAAC;QAED,OAAO;YACL,GAAG,WAAW;YACd,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;YAC9B,OAAO,EAAE,WAAW;SACrB,CAAC;IACJ,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAI,kBAAkB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAClC,sBAAsB,CAAC,UAAU,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE,EAAE,kBAAkB,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;QAChG,CAAC;QACD,iBAAiB,CAAC,OAAO,EAAE,UAAU,CAAC,CAAC;QACvC,MAAM,KAAK,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC;QAC1B,OAAO;YACL,QAAQ,EAAE,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAC,IAAI;YAC3C,QAAQ,EAAE,OAAO,CAAC,QAAQ,CAAC,QAAQ;YACnC,MAAM,EAAE,OAAO;YACf,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;YAC9B,OAAO,EAAE,IAAI;YACb,UAAU;YACV,WAAW,EAAE,OAAO,CAAC,WAAW;YAChC,KAAK,EAAE,OAAO,CAAC,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK;SACjE,CAAC;IACJ,CAAC;AACH,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"verdict-retry-prompt.d.ts","sourceRoot":"","sources":["../../src/agent/verdict-retry-prompt.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,sBAAsB,CAAC;AAErD,wBAAgB,uBAAuB,CAAC,QAAQ,EAAE,QAAQ,GAAG,MAAM,CAiBlE"}
|
|
@@ -0,0 +1,18 @@
|
|
|
1
|
+
export function buildVerdictRetryPrompt(scenario) {
|
|
2
|
+
const list = scenario.then
|
|
3
|
+
.map((assertion, i) => `${i + 1}. **${assertion}**`)
|
|
4
|
+
.join("\n");
|
|
5
|
+
return `## Verdict required
|
|
6
|
+
|
|
7
|
+
All Steps are complete (or the harness is ending this run). Emit your **final JSON verdict** now.
|
|
8
|
+
|
|
9
|
+
### Then checkpoints (${scenario.then.length} total — include each exactly once)
|
|
10
|
+
${list}
|
|
11
|
+
|
|
12
|
+
### Rules (strict)
|
|
13
|
+
- Do **not** use bash or read tools unless you lack CLI evidence for a checkpoint you already verified in this run.
|
|
14
|
+
- Your reply must include a fenced \`\`\`json block with \`status\`, \`checkpoints\`, and \`summary\` per the system prompt schema.
|
|
15
|
+
- \`status\` is \`"pass"\` only if every checkpoint passes with concrete CLI evidence.
|
|
16
|
+
- Map every Then bullet above to one \`checkpoints[]\` entry using the same assertion text.`;
|
|
17
|
+
}
|
|
18
|
+
//# sourceMappingURL=verdict-retry-prompt.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"verdict-retry-prompt.js","sourceRoot":"","sources":["../../src/agent/verdict-retry-prompt.ts"],"names":[],"mappings":"AAEA,MAAM,UAAU,uBAAuB,CAAC,QAAkB;IACxD,MAAM,IAAI,GAAG,QAAQ,CAAC,IAAI;SACvB,GAAG,CAAC,CAAC,SAAS,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,OAAO,SAAS,IAAI,CAAC;SACnD,IAAI,CAAC,IAAI,CAAC,CAAC;IAEd,OAAO;;;;wBAIe,QAAQ,CAAC,IAAI,CAAC,MAAM;EAC1C,IAAI;;;;;;4FAMsF,CAAC;AAC7F,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"verdict-retry-prompt.test.d.ts","sourceRoot":"","sources":["../../src/agent/verdict-retry-prompt.test.ts"],"names":[],"mappings":""}
|