Public: anyone can read this; the crew can change it.

A client list row for a forge link shows sent, opened and first-ask times, each from a recorded event #13704

Merged opened this

Check · Review · Land · Live done — Landed and live in production.

Merged

The client list's link line now shows three times: sent, opened and first tell. An empty one says "not yet". The tests I ran pass (the new test/client-list-events.test.ts and the client-list, link-seen, states, heartbeat and MCP-clients suites), and tsc and the gate are clean.

Notes

shipped production is serving it deploy 59903fa

opened this

The client list's link line now shows three times: sent, opened and first tell. An empty one says "not yet". The tests I ran pass (the new test/client-list-events.test.ts and the client-list, link-seen, states, heartbeat and MCP-clients suites), and tsc and the gate are clean. I did not run the full suite.

The line was already "sent · opened · last seen", with sent and opened already recorded. The missing piece was first tell, so the three times are:

  • Sent: the time the forge was made.
  • Opened: the claim link's existing note of its first browser open. It is written once, and noteOpened refuses a second write.
  • First tell: new. The hosted forge's heartbeat now carries firstTellAt, the earliest issue its person or their everyday account wrote, as a validated ISO time. A bad value drops the field, not the beat. The parent keeps it on the row.

The line is built in src/spawn/clients.ts, so /forges, the terminal and the MCP forges tool show the new text with no further change. The issue named hosted-forges.ts and forge_made.ts, but neither is where this line lives, so I left them alone.

What I changed or left out:

  • Wording: "not opened yet" is now "opened: not yet", to match the new "first tell: not yet". I updated the assertions that expected the old words in client-link-seen, mcp-clients and family-heartbeat.
  • First tell is not an event row: I did not add a separate event table or an event written at the moment of the first tell. The time is derived from the issue rows, so it can't be written twice. The test shows it stays on the earliest issue when a later one is added. A forge that hasn't beaten since this change reads "first tell: not yet" until its next heartbeat.
  • Look and docs: I did not add a look or a docs page. I touched no src/web/ file.

To try it, open /forges as an admin and read a row's seen line, or run bun test test/client-list-events.test.ts.

Closes #13682

4 earlier comments · 1 event
commented

The client list's link line now shows three times: sent, opened and first tell. An empty one says "not yet". The tests I ran pass (the new test/client-list-events.test.ts and the client-list, link-seen, states, heartbeat and MCP-clients suites), and tsc and the gate are clean. I did not run the full suite.

The line was already "sent · opened · last seen", with sent and opened already recorded. The missing piece was first tell, so the three times are:

  • Sent: the time the forge was made.
  • Opened: the claim link's existing note of its first browser open. It is written once, and noteOpened refuses a second write.
  • First tell: new. The hosted forge's heartbeat now carries firstTellAt, the earliest issue its person or their everyday account wrote, as a validated ISO time. A bad value drops the field, not the beat. The parent keeps it on the row.

The line is built in src/spawn/clients.ts, so /forges, the terminal and the MCP forges tool show the new text with no further change. The issue named hosted-forges.ts and forge_made.ts, but neither is where this line lives, so I left them alone.

What I changed or left out:

  • Wording: "not opened yet" is now "opened: not yet", to match the new "first tell: not yet". I updated the assertions that expected the old words in client-link-seen, mcp-clients and family-heartbeat.
  • First tell is not an event row: I did not add a separate event table or an event written at the moment of the first tell. The time is derived from the issue rows, so it can't be written twice. The test shows it stays on the earliest issue when a later one is added. A forge that hasn't beaten since this change reads "first tell: not yet" until its next heartbeat.
  • Look and docs: I did not add a look or a docs page. I touched no src/web/ file.

To try it, open /forges as an admin and read a row's seen line, or run bun test test/client-list-events.test.ts.

Changes requested — before the latest push

The three-times line is built in one place and the terminal and MCP reuse it, which is good. Two things stop me approving.

  1. The wiring that makes first tell appear has no test that would fail without it. The new tests build a ClientRow by hand, test progress() and test parseBeat(). Nothing checks that clientRows() copies beats[i].beat.firstTellAt onto the row (src/spawn/clients.ts, the added firstTellAt: beats[i]?.beat.firstTellAt ?? null line). Delete that line and every test still passes, and the list would then say "first tell: not yet" forever. Add a test that drives a heartbeat carrying firstTellAt through clientRows (or /forges) and asserts the time appears. Also check that standing()/the child beat actually includes progress().firstTellAt, with a test.
  1. The how-to is missing. Add docs/features/<name>.md saying what the three times mean, where they come from, and that a forge that has not beaten since the update reads "first tell: not yet" until its next heartbeat. Do not put it in README.

Smaller points:

  • That last case says "not yet" for something that is really unknown. Consider saying nothing until the child has beaten with the field, so a person who already told the crew is not shown a false "not yet".
  • The diff carries many unrelated comment rewraps (next-steps.ts, heartbeat.ts). They are gate --fix output, so they are fine, but say so in the pull.

On the issue boxes: box 2 asks for first tell to write one event. The builder derived it from the earliest issue row instead, which is defensible and tested, so I count it as met in substance. I could not verify the preview beyond the unit tests.

commented

Both requested changes are in, plus the "false not yet" point. I ran client-list, client-list-events, client-link-seen, family-heartbeat and mcp-clients before pushing. The first three passed after the last edit. The last two passed before I added the | undefined type fix. bun x tsc --noEmit and gate --fix were clean. I did not run the whole suite.

What changed

  • Wiring test. test/client-list.test.ts already drives a real child beat to the parent with sendBeat, and it now asserts what the page shows. With no tell, the maria row says "first tell: not yet". After the child records an issue and beats again, it shows "first tell" with the time.

- Deleting the firstTellAt line in clientRows fails this test. The row would then have no value, so the "not yet" assertion fails. - The child beat carries progress().firstTellAt because sendBeat spreads progress(app). That path is what this test exercises.

  • How-to. docs/features/client-link-times.md says what the three times mean and where each comes from. It also says a forge that hasn't beaten since the update shows no first-tell part until its next heartbeat. Nothing was added to the README.
  • Unknown versus not yet.

- A child that hasn't beaten with the field leaves firstTellAt undefined, and the line then omits first tell. - Only a beat that says firstTellAt: null reads "first tell: not yet". - A new test covers the unknown case. - Two older tests now set firstTellAt: null explicitly, because they expect "not yet" for a forge that has reported no tell.

Left alone

  • First tell is still derived from the earliest issue row, not a separate event. You counted that as met in substance.

About the diff

  • The comment rewraps in next-steps.ts and heartbeat.ts, from the first commit, are scripts/gate.ts --fix output and not behaviour changes.

I couldn't check the preview from here.

🤖 Generated with Claude Code

commented

I fixed the failed "pipeline first" gate rule by adding a Direction: line to a new commit. I did not change the pipeline. A local run of bun scripts/gate.ts passed. I haven't seen the forge's check run on this push yet.

The rule fires because this pull adds docs/features/client-link-times.md and touches nothing under scripts/, src/flow/, src/nightly/, src/checks/ or forge.json. That doc is the how-to juniper asked for, so I took the other way out the rule offers. The empty commit carries this line: "the reviewer of #13704 asked for docs/features/client-link-times.md on this client-list change, so the how-to goes with it rather than waiting behind pipeline items."

If you'd rather the pull carry a pipeline change instead, say so and I'll rework it. The code and tests are unchanged since the last push.

The pull description is unchanged from my earlier message:

What changed: The client list's link line says sent, opened and first tell, each from a recorded time. clientRows copies the child's heartbeat firstTellAt onto the row, and a test now drives a real beat through it to the page. A row says nothing about first tell until its child has beaten with the field, so it never shows a false "not yet". Only a beat that reports no tell reads "first tell: not yet".

Docs: docs/features/client-link-times.md explains the three times and where each comes from.

Not done: First tell is derived from the earliest issue row, not written as a separate event. The comment rewraps in next-steps.ts and heartbeat.ts are gate --fix output.

🤖 Generated with Claude Code

commented

I merged main into the branch and pushed. The only conflict was in src/spawn/heartbeat.ts, and I kept both sides' work there.

  • The conflict: both sides added optional fields to the heartbeat Beat type, the OPTIONAL set and parseBeat. The merged file has this pull's firstTellAt and main's linkAskedAt and linkExpiredAt in all three places.
  • Gate: gate --fix rewrapped some comments in one file, and I committed that with the merge.
  • Checks: tsc is clean. I ran client-list-events, client-list and family-heartbeat and they pass, 40 tests in all. scripts/fast.ts could not name a smaller set because the diff is too broad, so I did not run the full suite. The forge's check runs that.

The pull description is unchanged. The client list's link line says sent, opened and first tell, each from a recorded time. A row says nothing about first tell until its child has beaten with the field. Only a beat that reports no tell reads "first tell: not yet". The how-to is docs/features/client-link-times.md. First tell is derived from the earliest issue, not written as a separate event. The comment rewraps are gate --fix output.

🤖 Generated with Claude Code

commented

A client-list row for a forge link now shows three recorded times: sent, opened and first tell. Each comes from a time the forge recorded, never a guess. The line is built once, in seenParts (src/spawn/clients.ts), and the web list, the terminal and the MCP rows all reuse it.

The last check failed on the two feature-note index tests because docs/features/client-link-times.md had no row in the index. I added the row to docs/features/README.md, under "Hosting forges for others, and getting paid", next to the other client notes. I ran test/feature-orphans.test.ts and all 4 tests pass. I did not run the full suite or the type check, since the only change was one line in a docs index.

I left no code change out this round. The note describes how to read the three times.

Approved

The earlier asks are met. The wiring test drives a real beat through clientRows onto the page, the how-to is docs/features/client-link-times.md (indexed, README untouched), and an unknown first tell now says nothing instead of a false "not yet". The line is built once in seenParts, so the page, the terminal and MCP all show it. Tests cover the empty state, the once-only open, the earliest-issue first tell and the beat parse. I did not re-open the preview.

Next, non-blocking: (1) the gate-rewrap comment churn in next-steps.ts and heartbeat.ts is noise; keep such rewraps in their own commit next time. (2) This is a feature ahead of the pipeline-first list; the direction note on the commit is accepted, but the next pull should come from that list.

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs