Reasoning Chain Viewer Panel #18

New Issue

replit · 2026-03-20T22:25:27Z

replit commented

2026-03-20 22:25:27 +00:00

What & Why

Visitors pay for Timmy to think on their behalf. Showing Timmy's actual reasoning — the internal steps and considerations before the final answer — is both valuable and philosophically aligned with the project's transparency ethos. Pulls back the curtain on the AI process.

Done looks like

Each completed job in the Workshop has a 'Show Timmy's reasoning' expandable section
Reasoning captured server-side during job execution: Timmy prompted to produce chain-of-thought before final answer
Chain-of-thought stored separately from the final answer in jobs table
Clicking 'Show reasoning' expands a styled panel showing numbered steps
Panel clearly distinguishes reasoning (internal) from the final answer (delivered)
Captured for all paid jobs; not shown for free chat

Out of scope

True extended thinking API support (use explicit CoT prompting if not available)
Editing or replaying reasoning
Showing reasoning during streaming (shown only after completion)

Tasks

Chain-of-thought prompt — Update work model system prompt to produce a structured chain-of-thought before the final answer, delimited with clear markers for parsing.
Reasoning extraction + storage — Parse reasoning section out of model response; store in reasoning_chain column on jobs table; deliver final answer without reasoning prefix to visitor.
Reasoning API endpoint — Expose GET /api/jobs/:id/reasoning returning the reasoning chain for a completed job.
Reasoning viewer UI — Expandable 'Show Timmy's reasoning' section for each completed job in Workshop chat; styled distinctly from the answer.

Relevant files

artifacts/api-server/src/lib/agent.ts
artifacts/api-server/src/routes/jobs.ts
the-matrix/js/ui.js
the-matrix/js/session.js

## What & Why Visitors pay for Timmy to think on their behalf. Showing Timmy's actual reasoning — the internal steps and considerations before the final answer — is both valuable and philosophically aligned with the project's transparency ethos. Pulls back the curtain on the AI process. ## Done looks like - Each completed job in the Workshop has a 'Show Timmy's reasoning' expandable section - Reasoning captured server-side during job execution: Timmy prompted to produce chain-of-thought before final answer - Chain-of-thought stored separately from the final answer in jobs table - Clicking 'Show reasoning' expands a styled panel showing numbered steps - Panel clearly distinguishes reasoning (internal) from the final answer (delivered) - Captured for all paid jobs; not shown for free chat ## Out of scope - True extended thinking API support (use explicit CoT prompting if not available) - Editing or replaying reasoning - Showing reasoning during streaming (shown only after completion) ## Tasks 1. **Chain-of-thought prompt** — Update work model system prompt to produce a structured chain-of-thought before the final answer, delimited with clear markers for parsing. 2. **Reasoning extraction + storage** — Parse reasoning section out of model response; store in `reasoning_chain` column on jobs table; deliver final answer without reasoning prefix to visitor. 3. **Reasoning API endpoint** — Expose GET `/api/jobs/:id/reasoning` returning the reasoning chain for a completed job. 4. **Reasoning viewer UI** — Expandable 'Show Timmy's reasoning' section for each completed job in Workshop chat; styled distinctly from the answer. ## Relevant files - `artifacts/api-server/src/lib/agent.ts` - `artifacts/api-server/src/routes/jobs.ts` - `the-matrix/js/ui.js` - `the-matrix/js/session.js`

replit added the ai backend workshop frontend labels 2026-03-20 22:25:27 +00:00

gemini was assigned by Rockachopa

2026-03-22 23:37:40 +00:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: replit/timmy-tower#18