Skip to content

Point new users at one recommendation before a full report - #3023

Merged
HypForge merged 1 commit into
masterfrom
one-recommendation-first
Nov 18, 2026
Merged

HypForge merged 1 commit into
masterfrom
one-recommendation-first

Conversation

@bgmcmullen

Copy link
Copy Markdown
Contributor

New users were pushed toward a full report too early. The end of hyp setup and the README both named hyp report generate as the next thing to do. Both now lead with the one-recommendation ask from get one recommendation and fix it, and the report stays as the step for reviewing a whole period.

End of hyp setup

Before:

To draft a report from your recorded sessions, run `hyp report generate`.

After:

To get one recommendation from your recorded sessions, run:
  hyp ask "From my HypAware history, give me one recommendation that would most improve how my agents work this week."

The wording is the workflow page's "let it find the biggest one" prompt without "in this repo", because setup is not run from a project.

The gate follows the command it names. It still needs a client this run left recorded with its CLI on PATH, and now checks for the hypaware-query skill on disk. The hypaware-report skill and the hypaware-analyst worker no longer decide it. recordedReportClientLaunchable is renamed recordedAskClientLaunchable, and the test-only option reportGenerateAvailable becomes recommendAskAvailable. Placement is unchanged: ahead of the privacy close and the closing offers.

This reverses what #2704 (issue #418) added: setup no longer names hyp report generate at all.

README

"Reports" becomes "Recommendations and reports": the one-recommendation hyp ask command and a link to the workflow page come first, then the existing report command and publishing text.

Companion: hyparam/hypaware.ai#143 makes the same change to step 5 of the site quickstart.

Not in this PR

  • The analyst-worker check went away with the old gate, along with its test (a Claude install whose analyst copy failed is not ready). hyp report generate itself still checks for the report skill through reportSkillLaunchers, which is untouched.
  • Setup's closing offer still says "Run hyp ask any time: HypAware suggests a skill", so setup now names two hyp ask forms. Left as is.

Checks. npm test passes (9737 pass, 0 fail, 9 skipped) and npm run typecheck is clean. I did not run the wizard smokes or a real hyp setup.

CPU and memory. No concern. The gate runs once at the end of setup and does one fs.access per launchable recorded client, which is less work than the stat-per-client plus skill lookup it replaces.

🤖 Generated with Claude Code

The end of `hyp setup` and the README both sent people straight to
`hyp report generate`. A full report is the long way to a first fix.
Both now lead with the one-recommendation ask from
docs/workflows/get-one-recommendation.md.

Setup: the closing pointer prints the `hyp ask` command for one
recommendation instead of naming `hyp report generate`. Its gate follows
the command: a client this run left recorded, its CLI on PATH, and the
hypaware-query skill on disk. The report skill and the analyst worker no
longer decide it, so `recordedReportClientLaunchable` becomes
`recordedAskClientLaunchable`.

README: the Reports section becomes "Recommendations and reports", with
the one-recommendation ask first and the report as the step for
reviewing a whole period.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
@bgmcmullen bgmcmullen added the neutral:adopt Foreign PR adopted into neutral's reconcile scope label Nov 17, 2026
@HypForge HypForge added the neutral:adopted Adoption completion record: merged while carrying neutral:adopt (LLP 0031) label Nov 17, 2026
@HypForge

Copy link
Copy Markdown
Contributor

Final review (HypForge steward), head b86f308

Clean. Rating LOW.

What I checked

  • The gate. recordedAskClientLaunchable is the same check hyp ask uses for its paste fallback (recordedClientReadsQuerySkill): a client this run recorded, its CLI on PATH, and its hypaware-query skill on disk. So the pointer and hyp ask agree on who can answer.
    • Desktop-only, OpenCode and Pi, and dry, scripted or cancelled runs print no line.
    • The removed tests covered the analyst worker, which only hyp report generate needs. That command keeps its own skill check.
  • The command.
    • It has no shell-special characters, so it is safe as typed in bash, zsh, PowerShell and cmd.
    • It is 121 characters, so it soft-wraps at 80 and still copies as one line.
    • It matches the workflow page's "biggest one" prompt, minus "in this repo".
  • A real hyp ask with this prompt, in a temp home with a stub Claude Code, started the client in the current directory with the framed question and exited 0. I did not run a real agent answer, because that needs a paid account.
  • Reversal of Fix #418: setup names hyp report generate when a recorded client can start it #2704. No LLP requires setup to name hyp report generate, so nothing settled is undone.
  • README. The section matches the code and the workflow page, and its links resolve.
  • Tests.
    • The three walkthrough smokes pass.
    • The wizard, ask and report tests pass 645/645 locally.
    • All CI is green, including test-windows.
    • I did not run an interactive hyp setup in hyp-sandbox: it needs a driven TTY, and its first look can clash on ports with this host's daemon.

Lines

  • CPU and memory: no concern. The gate runs once at the end of setup, with one PATH lookup and one file check per recorded client.
  • Privacy: LOW (hypforge LLP 0015 scale). There is no new data, store or reader.
  • Security: no new spawn or input path.
  • Server-suite gate: not triggered. There is no CLI command name, contract, MCP codec or tool schema change.

For Brendan to know

  • The gate does not look at whether anything was recorded. A first setup with no imported history still prints "To get one recommendation from your recorded sessions". The question form then opens the agent on an empty history. The old report pointer behaved the same, so this is not new, and it is your call whether to also require recorded rows.
  • Setup naming two hyp ask forms is your recorded choice, so I left it.
  • The commit carries a Co-Authored-By trailer. A default squash would copy it onto master, so the squash body should leave it out.

@HypForge

Copy link
Copy Markdown
Contributor

Landing note (HypForge orchestrator): enqueuing at b86f308. The author handed this PR over with neutral:adopt at 23:16Z on 2026-11-17. It lands on a clean LOW review with green checks, so there is no merge ask.

  • Review: LOW, clean, at this exact head (final comment above, 6964606265). Privacy LOW. Security: no new spawn or input path. CPU and memory: no concern. Server-suite gate: not triggered.
  • HypForge pushed nothing to this branch.
  • Guard: open, not a draft, CLEAN, all checks green including test-windows, 0 behind master 4c304b9, master CI green, no release PR open.
  • Not run: an interactive hyp setup in the sandbox, and a real agent answer to the printed question.

One point left as it is unless the author says otherwise: the pointer prints even when nothing has been recorded yet, as the old report pointer did.

A merge is not a release. The new setup line reaches users with the next published version. The companion site PR (hyparam/hypaware.ai#143) is held for a separate reason, stated on that PR.

@HypForge
HypForge added this pull request to the merge queue Nov 17, 2026
Merged via the queue into master with commit 1720cb1 Nov 18, 2026
10 checks passed
@HypForge
HypForge deleted the one-recommendation-first branch November 18, 2026 00:03
@HypForge

Copy link
Copy Markdown
Contributor

Merged (HypForge orchestrator) as 1720cb1 at 00:03:56Z on 2026-11-18, on the first pass through the merge queue. The merge-group CI was green. The push CI on this commit was cancelled when the next merge (#3025) arrived a few minutes later, as the queue does; CI on that newer commit is running.

  • This is on master only. The new closing line of hyp setup reaches users with the next client release.
  • The companion site PR (hyparam/hypaware.ai#143) is held for a separate reason, stated on that PR: a site deploy now would publish docs for a command the released client does not have.
  • Not run: an interactive hyp setup in the sandbox, and a real agent answer to the printed question.
  • Shorten the CLI: top-level fix, purge and skills, a recommendation group, direct query graph and vector #3024 touches the same README and wizard files; a test merge of the two was conflict-free.

@HypForge HypForge mentioned this pull request Nov 18, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

neutral:adopt Foreign PR adopted into neutral's reconcile scope neutral:adopted Adoption completion record: merged while carrying neutral:adopt (LLP 0031)

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants