Skip to content

docs(skills): add staleness check + low-resolution issue template - #44

Merged
tato123 merged 1 commit into
mainfrom
docs/skill-staleness-and-low-resolution-template
Apr 27, 2026
Merged

tato123 merged 1 commit into
mainfrom
docs/skill-staleness-and-low-resolution-template

Conversation

@tato123

@tato123 tato123 commented Apr 27, 2026

Copy link
Copy Markdown
Contributor

Summary

  • amos-next: new Step 2.5 "Verify against current state (staleness check)" — picker audits the issue body against current code + current issue state before announcing a plan, updates the body in place with strike-throughs when load-bearing claims have gone stale, and announces a fresh plan that supersedes the body where evidence has shifted. Step 3 announcement, Step 5 scope, and Step 7 pre-PR review gate updated to match.
  • amos-file: default template rewritten to be low-resolution. Issues capture goals, not pre-implementation plans; the picker produces the high-resolution plan at pickup time. AI Agent Notes default to "None." with explicit examples of what does and does not belong (search-fuel + ruled-out approaches with reasoning belong; ordering suggestions and file paths don't).

Why

streamlib issue #528 surfaced repeated staleness in issue bodies — impl issues already filed, dependency edges already wired, "defects" already fixed at compile time. The pickup-research loop catches that automatically, and the lower-resolution template prevents new issues from being filed with the same staleness problem baked in. See tatolab/streamlib#540 for the matching streamlib-side changes (CLAUDE.md "Editing markdown documentation" rules, "Issues are goals, not specs" reframing in Work Tracking, lower-resolution docs/issue-template.md).

Test plan

  • Both files were already in active use locally for the streamlib #528 work — the staleness check ran successfully (caught three stale AI Agent Notes in the original issue body).
  • No code in this PR — pure skill-doc updates.

Follow-ups

None.

🤖 Generated with Claude Code

Mirrors two improvements that landed in ~/.claude/skills/ during
recent streamlib work back into the canonical repo copies:

amos-next:
- New Step 2.5 "Verify against current state (staleness check)" that
  requires the picker to audit the issue body against current code +
  current issue state before announcing a plan. Picker updates the
  issue body in place (with strike-throughs and reasoning preserved)
  when load-bearing claims have gone stale, then announces a fresh
  plan that supersedes the body where evidence has shifted.
- Step 3 announcement template revised to lead with "Goal" +
  "Refined plan (after Step 2.5 staleness check)" + "What changed
  vs. issue body" rather than treating the issue body as
  authoritative.
- Step 5 reframes scope as "the goal as refined in Step 3's plan,"
  not a literal reading of every checkbox in the issue body.
- Also pulls in the pre-PR review gate (Step 7) that was added in
  ~/.claude but hadn't been mirrored back yet — same content already
  in use locally, see streamlib PR #540.

amos-file:
- "Amos default template" rewritten to be low-resolution by default.
  Issues capture goals not pre-implementation plans; the picker
  produces the high-resolution plan at pickup time. Specifics in the
  issue body just create staleness for the next agent to clean up.
- AI Agent Notes default to "None." with examples of content that
  does and does not belong (search-fuel and ruled-out approaches
  with reasoning belong; ordering suggestions and file paths don't —
  they go stale fast).

Why: streamlib issue #528 surfaced repeated staleness in issue bodies
(impl issues already filed, dependency edges already wired, "defects"
already fixed at compile time). The pickup-research loop catches
that automatically; the lower-resolution template prevents new
issues from being filed with the same staleness problem baked in.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@tato123
tato123 merged commit 606973e into main Apr 27, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant