fix(slack-bot): require structured classifier output

[ColeMurray]

Summary

• require Slack repo classification responses to come via Anthropic tool_use (classify_repository)
• remove heuristic keyword fallback and legacy text-JSON parsing
• return low-confidence clarification when structured output is missing or invalid
• add classifier tests for valid structured output, invalid tool payload, and missing tool output

Validation

• npm run test -w @open-inspect/slack-bot -- --run src/classifier/index.test.ts
• npm run build -w @open-inspect/shared
• npm run typecheck -w @open-inspect/slack-bot

[github-actions]

Terraform Validation Results

Step	Status
Format	⚠️
Init	✅
Validate	✅

Note: Terraform plan was skipped because secrets are not configured. This is expected for external contributors. See docs/GETTING_STARTED.md for setup instructions.

Pushed by: @ColeMurray, Action: pull_request

[greptile-apps]

Greptile Overview

Greptile Summary

• Switched Slack repo classification to require Anthropic tool_use structured output (classify_repository) instead of parsing JSON from text.
• Added normalization/validation for tool payloads and a deterministic model call (temperature: 0, forced tool choice).
• Removed heuristic keyword fallback; on invalid/missing tool output the classifier now returns a low-confidence clarification with repo alternatives.
• Added unit tests covering valid tool output, invalid tool payload, and missing tool output paths.

[greptile-apps]

_{2 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

Additional Comments (1)

packages/slack-bot/src/classifier/index.ts
Medium confidence auto-clarifies

needsClarification currently becomes true for confidence === "medium" only if alternatives.length > 0. With the new tool schema, the model can return confidence: "medium" with an empty alternatives array, which will make the bot proceed without asking the user even though the model explicitly said it’s not confident. This is a reachable behavior regression. Consider treating any non-high confidence as requiring clarification (or enforce non-empty alternatives when confidence != high).

Prompt To Fix With AI

This is a comment left during a code review.
Path: packages/slack-bot/src/classifier/index.ts
Line: 279:287

Comment:
**Medium confidence auto-clarifies**

`needsClarification` currently becomes `true` for `confidence === "medium"` *only if* `alternatives.length > 0`. With the new tool schema, the model can return `confidence: "medium"` with an empty `alternatives` array, which will make the bot proceed without asking the user even though the model explicitly said it’s not confident. This is a reachable behavior regression. Consider treating any non-`high` confidence as requiring clarification (or enforce non-empty `alternatives` when confidence != high).

How can I resolve this? If you propose a fix, please make it concise.