Skip to content

emrg: the Feishu task's report does not say what a conclusion rests on #2670

Closed

Description

@argszero

emrg/server/feishu_prompt.md §4 (Report) tells a round what to close with — how many messages arrived per verdict, what it did, which requests are waiting for the host, and what is blocked — and says nothing about the evidence behind a conclusion about how something ran. So a round can report "the worker crashed at step 3" or "the request was rejected upstream" without naming the file it read or an address a reader could open, and nothing in the template asks for it.

What is observed. On master at a14727e9, §4 is two paragraphs: one listing what the report contains, one requiring a pending request to name the Feishu message id and conversation it was asked in. Neither mentions a log, a file, or an address. The Forbidden list's only related line is that a credential is never logged.

Why it matters. The report is the artifact the host reads; the round's own working state is not. A conclusion with no address cannot be checked by the person acting on it, and an unverifiable conclusion is indistinguishable from a guess. The failure has a second half: when a file cannot be retrieved, saying nothing about it reads exactly like having read it and found nothing of interest — so the one case where the round is blind is the case that looks cleanest.

What the fix must satisfy.

  1. §4 states that a conclusion about how something ran names the file it was read from, and an address the reader can open for it.
  2. One address per file involved: a conclusion drawn from three files is three readings, and a reader who can open only one cannot tell which of them said it.
  3. A file the round could not retrieve is named with the step it could not pass, rather than left out.
  4. Written in the template's own voice, in §4 where the report is defined; no host attribution, no dated history, and no installation-specific address form — the shape of the address belongs to the installation, the obligation belongs here.
  5. A test naming the rule, so deleting it reddens — asserted both on the file and on the render, since a template construct can keep every character and send none of it (tests/test_feishu_prompt.py already renders through the real _build_evolution_prompt).

Activity

  1. argszero commented on Nov 2, 2026

    @argszero
    OwnerAuthor

    Handled by #2671

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions