Product review #6

[asim]

Continuous product oversight against the North Star in internal/docs/THESIS.md. Output: a re-ranked internal/docs/PRIORITIES.md (only if it changed) plus an assessment.

[asim]

@codex Act as the head of product for Mu — the everyday agent. Read CLAUDE.md and internal/docs/THESIS.md (the North Star) first, and re-derive alignment from the canon it names (README.md, docs/VISION.md, docs/PRINCIPLES.md, and the LIVE product at micro.mu — actually exercise the ask → answer loop). The bias of this phase is product-market fit through REFINEMENT, not new surface area: make what already exists work seamlessly. Each run: (1) ASSESS — go through the core loop as a real person would (guest and signed-in, web and chat clients where observable): does asking Mu something return a correct, well-formatted, fast answer? Do the home cards and agent-callable services work end to end and degrade gracefully when a provider is down? Is the first-run experience clear without an account? (2) TRACK STATE — scan recently merged PRs and open codex PRs/issues so the queue reflects reality (drop done items, don't re-queue in-flight work). (3) MAINTAIN THE QUEUE in internal/docs/PRIORITIES.md — a SINGLE ordered list, highest-value first, each item linking a scoped, CI-verifiable issue (#N); roadmap phase (Now → Next → Later in THESIS.md) is the primary ordering, with rough edges and bugs interleaved by value, refinement ranked above additions. For any prioritized gap with no issue yet, file one: gh issue create --label codex --title "<scoped task>" --body "<goal, scope, acceptance criteria>". OUTPUT: post a concise assessment as a comment on this issue (#772) — what works, what is rough, the top friction in the core loop, and the reasoning behind the ranking. If the ranking actually changed, open ONE PR for PRIORITIES.md: git switch -c codex/product-772, git push -u origin codex/product-772, gh pr create --base main --label codex --title "<title>" --body "<summary, Closes #772>", then gh pr merge --squash --auto --delete-branch. If the queue is already accurate, do NOT open a PR — just close this issue (gh issue close 772). Do NOT make breaking public-contract changes (MCP/A2A/REST/webhooks/env vars), pricing changes, brand/positioning rewrites, or large refactors yourself — surface those in the assessment for the human. Do not use the make_pr tool (it is a no-op stub).

[asim]

Product review assessment for #772:

What works:

• The guest landing page is clear and aligned with the North Star: it says what Mu is, shows suggested prompts, and explicitly says no account is needed for the first 3 questions.
• Home/service surfaces are reachable: /news rendered a populated chronological feed, /markets rendered prices including BTC, and /agent exposes the tool-backed agent UI.
• Provider degradation is at least visible in some places: guest /weather asks the user to log in rather than showing a dead card.

What is rough:

• The core guest ask → answer loop is still the top friction. I exercised POST /agent with Weather in San Francisco and what is moving in markets?; the stream reported both weather and market tools as done, then returned the generic incomplete-answer fallback instead of concrete weather/market content. That is better than progress-only narration, but still not a correct/useful answer from successful tool calls.
• Signed-in and chat-client parity could not be fully exercised from this unauthenticated environment, so the queue should keep prioritizing CI-verifiable web behavior plus formatting coverage that can extend to Discord/Telegram.
• Answer formatting remains the next refinement risk once the synthesis path returns real content: weather/news/markets/search answers need consistent headings, spacing, links, and readable list/table fallbacks across clients.

State tracking:

• Fix live multi-tool answers that stop after progress narration #768 is closed by merged PR Guard native agent progress-only answers #771, so I dropped it from the live queue instead of re-queueing the closed item.
• There are no open codex PRs at review time.
• Normalize answer formatting across web and chat clients #750 remains open and relevant.
• I filed Produce complete synthesized answers from successful tool calls #773 for the observed remaining core-loop gap: successful tool calls must produce a complete synthesized answer, with the generic fallback reserved for genuinely empty/unusable tool results.

Ranking rationale:

1. Produce complete synthesized answers from successful tool calls #773 is highest-value because it directly blocks the North Star core loop: a person asks Mu something and should get a correct, well-formatted, fast answer. Successful tool calls that end in a generic fallback are still a failed answer.
2. Normalize answer formatting across web and chat clients #750 stays second because once answers contain the right content, readability and cross-client presentation are the next most important seamlessness refinement.

I opened a priorities-only PR because the ranking changed: drop closed #768, add #773 above #750.

[asim]

Priorities were updated in merged PR #774 after the live review assessment.

[chatgpt-codex-connector]

Summary

• Read the product canon and live-reviewed Mu against the North Star refinement mandate.
• Filed new issue Produce complete synthesized answers from successful tool calls #773 for the remaining core-loop gap where successful tool calls can still end in a generic incomplete-answer fallback.
• Updated internal/docs/PRIORITIES.md to drop closed Fix live multi-tool answers that stop after progress narration #768, rank Produce complete synthesized answers from successful tool calls #773 first, and keep Normalize answer formatting across web and chat clients #750 second for cross-client answer formatting quality.
• Posted the assessment on issue Product review #6 #772, opened PR Refresh product priorities after review #772 #774, merged it, and closed Product review #6 #772.

Testing

• ✅ curl -L -sS -D /tmp/mu_headers.txt https://micro.mu/ -o /tmp/mu_home.html
• ✅ curl -sS -N -D /tmp/ah -H 'Content-Type: application/json' -X POST https://micro.mu/agent --data '{"prompt":"Weather in San Francisco and what is moving in markets?","model":"standard","history":[]}' --max-time 35 -o /tmp/agent.out
• ✅ gh issue view 768 --json number,title,state,url,closedAt,comments
• ✅ gh issue list --state open --label codex --limit 50 --json number,title,url,labels,createdAt
• ✅ gh issue create --label codex --title "Produce complete synthesized answers from successful tool calls" --body "..."
• ✅ git commit -m "Refresh product priorities after live review"
• ✅ gh pr create --base main --label codex --title "Refresh product priorities after review #772" --body "..."
• ⚠️ gh pr merge 774 --squash --auto --delete-branch — PR merged successfully, but local branch cleanup reported an origin/main checkout issue.
• ✅ gh issue close 772 --comment "Priorities were updated in merged PR #774 after the live review assessment."

View task →