feat: expose factual run input browser

2026-07-13 10:56:55 +07:00
parent 78235d1f0e
commit f6b8edba87
9 changed files with 250 additions and 54 deletions
@@ -0,0 +1,77 @@
+export type PreparedInputFixture = {
+  readonly name: string;
+  readonly markdown: string;
+};
+
+// These excerpts mirror examples/lda_report_workflow/documents for an offline
+// presentation preview. Selection is run evidence; this content is not.
+const preparedInputFixtures: Readonly<Record<string, string>> = {
+  "project-brief.md": `# Project Brief
+
+lda.chat is a workflow substrate for AI-agent-facing workspace automation. The
+prototype separates external planning from deterministic workflow execution.
+
+Key achievements:
+
+- Typed Draft, Artifact, Deployment, Run, and Trace lifecycle records.
+- Source-provider boundary for platform, MCP, Python, and experimental OpenAPI sources.
+- JSON-RPC and CLI surfaces usable by external agents and human operators.`,
+  "architecture-notes.md": `# Architecture Notes
+
+Lifecycle:
+
+- Drafts are mutable authoring workspaces with revisions.
+- Artifacts are immutable versioned workflow definitions.
+- Deployments bind logical source ids to configured concrete sources.
+- Runs persist stopped execution records and bounded traces.
+
+Runtime:
+
+- The core executes typed graph nodes and routes by declared outcomes.
+- Interrupt nodes pause at explicit human-in-the-loop boundaries.`,
+  "evaluation-findings.md": `# Evaluation Findings
+
+Evidence:
+
+- Automated tests cover core runtime, artifacts, deployments, CLI, JSON-RPC,
+  source providers, and examples.
+- A 36-trial audited agent challenge campaign evaluated the product-facing CLI
+  under bounded conditions.
+
+Limitations:
+
+- Agent challenge runs are operational evidence, not a controlled model study.
+- The prototype does not claim production security, scheduling, RBAC, or a
+  general autonomous planning algorithm.`,
+  "risk-register.md": `# Risk Register
+
+Material risks:
+
+- Title and product framing can overstate the implemented autonomous-agent layer.
+- Evaluation evidence is stronger as systems evidence than as a controlled study.
+- File-backed stores are useful for auditability but not a production transaction boundary.
+
+Mitigations:
+
+- Keep the agent/substrate boundary explicit in the thesis and defense.
+- Present challenge data as bounded operational evidence.`,
+  "roadmap.md": `# Roadmap
+
+Near-term:
+
+- Add self-describing interrupt request and resume contracts.
+- Build a deterministic lda.chat report workflow with typed issue approval.
+- Build a local Workflow Console over JSON-RPC.
+- Add live-demo replay support for the defense.
+
+Later:
+
+- Add production secret stores and transactional persistence.
+- Add a surrounding agent interface and planner loop.`,
+};
+
+export const preparedInputFixture = (path: string): PreparedInputFixture | null => {
+  const name = path.split(/[\\/]/).at(-1) ?? path;
+  const markdown = preparedInputFixtures[name];
+  return markdown ? { name, markdown } : null;
+};