Project tool observability into reviewer transcripts

[chubes4]

Summary

Project canonical Agents API tool lifecycle summaries into sanitized WP Codebox reviewer evidence.

What changed

• Added strict versioned normalization for ordered tool calls, safe identifiers, redacted argument key summaries, fixed error messages, and canonical result shapes.
• Removed raw tool observability before staging and records only the sanitized projection and digest in reviewer artifacts.

How to test

1. Run npm run build; expect TypeScript compilation and CLI executable validation succeed..
2. Run npm run test:agent-task-contracts; expect Agent-task, workflow interface, runtime materialization, privacy, malformed-input, and result-shape contracts pass..
3. Run npm run test:browser-runner-template; expect The generated browser runner template contract passes..

Compatibility

Additive reviewer-evidence behavior only. Existing agent-task result fields, artifact privacy rules, optional-artifact behavior, and no-op semantics are preserved. Unsupported or malformed observability is omitted.

Evidence

• CI expected: Repository pull request checks
• Reviewer-resolvable source evidence: Expose compact tool observability in chat metadata agents-api#428
• Reviewer-resolvable source evidence: Project Agents API tool observability into reviewer transcripts #1806
• agent-task-contracts: Passed
• browser-runner-template: Passed
• build: Passed

AI assistance

• AI assistance: Yes
• Tool(s): OpenCode
• Model: openai/gpt-5.6-terra
• Used for: Implemented the canonical observability projection, corrected result-shape handling, hardened identifier privacy, added regression coverage, and verified reviewer artifact staging; Chris reviewed and owns the change.

Source relationships

• Closes Project Agents API tool observability into reviewer transcripts #1806
• Relates to Review test adapter runs WP Codebox TypeScript tests with bare node Extra-Chill/homeboy#8324