goldfish — fresh-context review
SkillMediaReview a design artifact (spec, plan, PR packet) with fresh, memoryless subagents. Invoke when the user asks — "goldfish this", "fresh eyes", "tear this apart" — or after suggesting it for a spec that has grown load-bearing and the user agrees.
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the goldfish — fresh-context review skill
What this skill tells your AI
The instructions your AI receives, as published by zoltanmaric/coppersushi in .agents/skills/goldfish/SKILL.md and read by ahel’s review.
Intent: an artifact written in a context-rich session always reads complete to that session. The test is whether a reader with zero session context can act on it.
A goldfish is a fresh subagent given only the artifact plus what any fresh context can legitimately reach: the repo, the wiki, linked sources. Session context is out of bounds — don't paste in your intent or the answer you hope for.
Facets — a menu, not a mandate; pick by stakes. Run the deleter before critic and readiness: optimizing a thing that shouldn't exist is the classic mistake.
- Requirement deleter — whole before parts: should this task exist at all? Is it the right question? Only then: which individual requirements can be cut?
- Comprehension — can you explain the goal and the touched system from what you can read?
- Simplifier — of what survives, is there a simpler approach?
- Critic — assume defects exist; verify claims against reality (open the code, run the command), not against the artifact's account of it.
- Readiness — does this contain what the next action needs? Missing implementation detail is NOT a finding.
Sandbox (non-negotiable): reviewers never mutate a checkout they don't own. Read-only subagents by default; a reviewer that must build or run gets its own disposable worktree.
Exit: findings filtered by you — a finding survives if acting on it would change what gets built or what risk is visible. Weigh findings toward the next action and the direction's validity; defects in far-future steps are usually premature — the user works iteratively, and those steps will be re-planned before they're reached. Done when a fresh reader can act on the artifact without rediscovering intent. If substantive findings persist after a couple of rounds, the artifact needs rethinking, not more polish.
Signals
- GitHub stars
- 69
- Forks
- 11
- Last commit
- Sep 2026
Advanced
- Catalog kind
- skill
- Gateway key
goldfish- Source
- github.com/zoltanmaric/coppersushi