Austin Griffith 6fad7ae222 feedback: shift trigger from conditional to imperative (#157) ...

PR #156 added a footer pointer to feedback/SKILL.md at the bottom of
five high-traffic skills. A week later, zero feedback has come in.

The hypothesis: discovery wasn't the bottleneck — the trigger was.
"Found something wrong, confusing, or genuinely helpful?" requires
an agent to introspect on its own session and judge whether the
result clears a bar. The default resolution to that question is
"keep moving."

This shifts the framing from introspective opt-in to ritualized
end-of-task default:

- All five skill footers: "Before you finish: send a one-line note
  via feedback/SKILL.md — issue or praise." Same wording across
  every footer for consistency.
- ship/SKILL.md gets two extra surfaces: a checklist item in the
  Quick-Start Checklist and a "Wrap" row in the Skill Routing Table.
  ship/ is the routing skill agents read top-down at the start of
  every dApp build, so feedback now appears on the path, not just
  in the footer.
- "Issue or praise" replaces "wrong, confusing, or genuinely
  helpful." Shorter, more balanced, and explicitly invites praise
  (which the inbox is undersupplied on).

If this still produces no feedback after a couple weeks, the next
move is probably telemetry on /api/feedback hits — distinguishing
"agents try and get rejected" from "agents never call it" would
tell us whether the issue is reach or wiring.

Co-authored-by: austintgriffith <austin.griffith@ethereum.org>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-05-08 11:57:00 -06:00

..

SKILL.md

feedback: shift trigger from conditional to imperative (#157)

2026-05-08 11:57:00 -06:00