Files
plan2code/src/web-console/console-chat.md
T

4.8 KiB

Quick Questions (the Ask tab)

The console points here the first time a Quick question reaches you. The person can ask anything on the Ask tab at any time, about this project's files or in general, without waiting for a card. A message never blocks a card send and a card send never blocks a message.

How they arrive

wait exit 0 (or chat) carries them in reply, after the card lines, plus chat: the seq number of each question, the ids your replies need as re.

Quick question (12): Where does the export go?
  About: Export format
  Image: /home/sam/.plan2code/console/sessions/<sid>/uploads/3f2b9c1e.jpg (shot.png)
  File: /home/sam/.plan2code/console/sessions/<sid>/uploads/9e4b2d7a.json (config.json)
Declined (14): edit r7
  Note: keep the old name working too
Approved (15): edit r9
New conversation started. Treat earlier chat as closed.
  • Open every Image: and File: path with your file-read tool before you answer.
  • About: is set only when the person attached that context; without it the question is general, not about the card on screen.
  • After "New conversation started" treat earlier chat as closed and do not raise it unless they do.
  • node "<CONSOLE>" chat --session <sid> checks now and never sleeps: exit 0 with the same reply and chat, 10 nothing waiting, 20 server gone. It never reads card sends.

Answering

Every message gets a reply, whatever it asks and whatever you are mid-task on. post and wait print unansweredChat with a warning for each message not yet replied to: stop and reply. When a message asks you to do something, do it if you can, then say what happened; if you will not, say so and why.

  1. Handle any card answers first, under the workflow's own rules.
  2. Write the replies, in order.
  3. Send one post with the card updates and the replies.

{"chat":{"replies":[{"re":12,"md":"The export is written by src/export.js, into out/."}]}}

  • re is the message's seq; md is markdown and non-empty. id is optional (default r<re>): give one only for two replies to the same message. Replies merge by id, so re-posting the same reply replaces it.
  • Print one terminal line per exchange: Quick question: <first ~60 characters> (answered in the console).
  • Answer questions about the project from files and read-only lookups (search, git log, reading); run nothing else.
  • For what a Plan2Code skill does, asks or writes, read that skill's SKILL.md (~/.agents/skills/<skill>/SKILL.md, or the table "What each skill reads and writes" in ~/.agents/skills/plan2code/SKILL.md) first. Name files exactly as it does; if it does not say, say you do not know.
  • With About:, answer from that context.

A prompt to carry away

When the right answer is a separate task (a new Plan, Quick Task or another skill), explain first, then give ONE self-contained prompt in a fenced block with no info string (the page adds a Copy button). The block:

  • opens with plain prose, never a command word: a first line that looks like a shell command (git, npm, node, rm, #!, or an everyday verb like Make, Go, Source, Export, Cat, Echo, Zip, Tar, Cd, Ls; case-insensitive) gets a "paste into the Ask tab" hint. Start with "In this repo, ..." or "Using the Quick Task skill, ...". A leading /plan2code-quick-task ... is fine.
  • names the skill to paste it into, the files to read first, the task, and what is out of scope.
  • stands alone: it assumes nothing said in this session.

Never steer the workflow

A chat message never answers a card, passes a gate or changes where the workflow goes ("approve phase 2"): point to the card (its note box, its answer) and do not act. Chat is never recorded under specs/. A conversation holds at most 10 typed messages; the page enforces it.

Edits need approval

  • A request to change files gets a reply with proposal (a plain-English summary and every file that will change) and no change yet: {"chat":{"replies":[{"re":14,"md":"I can rename it.","proposal":{"summary":"Write the export to out/audit.csv","files":["src/export.js","README.md"]}}]}}. The page shows Approve and Decline.
  • Approved (<seq>): edit <id>: make exactly the proposed change, then reply (its re is that seq) with changed:[{file,added,removed}], line counts from git diff --numstat -- <files> (counted by hand for an untracked file). Name the changed files in your next update.
  • Declined (<seq>), with any note: acknowledge and change nothing; a revised proposal is a new reply.
  • Never touch the running skill's own files under specs/<idea>/ (a proposal naming them is refused). Run no commands beyond read-only lookups, except one they plainly ask for in their message (a commit they spell out, a test run): run it and report. Refuse anything else, saying why.
  • Edits happen at check-ins, never mid-task in a build.