Menü
Pattern for progressively refining context retrieval to solve the subagent context problem.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Basierend auf der SOC-Berufsklassifikation
Use when designing memory or persistent context for an agent — deciding what to persist vs. recompute, how to retrieve only what's relevant, and how to summarize/compact history to fit the context window. Use when an agent forgets, bloats its context, or repeats work.
Interactive installer for Rosetta. Guides users through selecting and installing skills and language-specific rules at user-level or project-level, verifies paths, and optionally optimizes installed files.
Instinct-based learning system that observes sessions via hooks, creates atomic instincts with confidence scoring, and evolves them into skills, commands, and agents.
Use when building or reviewing an evaluation for an LLM feature — assembling a representative test set, choosing pass criteria (exact match, programmatic checks, rubric, or LLM-as-judge), and catching regressions. Use when asking "how do I know this prompt or model change is better?"
Use when writing or reviewing a prompt for an LLM — making it specific, structured, and testable (clear task, role/context, examples, an explicit output contract, decomposition, room to reason) and avoiding the anti-patterns that make prompts brittle. Provider-agnostic.
Use when creating or structuring a Claude Code plugin for this marketplace — the on-disk layout, the plugin.json manifest fields, the frontmatter for skills/commands/agents, and the authoring rules (no absolute paths, least-privilege tools, one coherent capability).
| name | iterative-retrieval |
| description | Pattern for progressively refining context retrieval to solve the subagent context problem. |
Solves the subagent context problem: subagents don't know what files or terminology they need until they start working. Sending everything exceeds context limits; guessing is usually wrong.
A loop of up to 3 cycles, stopping early when context is sufficient:
Dispatch — broad initial query: glob patterns, keywords, excludes.
Evaluate — score each retrieved file for relevance to the task (0–1). Identify what context is still missing.
Refine — update the query based on findings: add terminology and patterns discovered in high-relevance files, exclude confirmed irrelevant paths, target identified gaps.
Repeat — run at most 3 cycles. Stop early when you have 3+ high-relevance files (≥0.7) and no critical gaps.
Start broad — the first cycle often reveals the codebase's actual terminology (e.g. "throttle" instead of "rate limit"), which drives better subsequent queries. Track gaps explicitly: knowing what's missing is what makes refinement useful. Three high-relevance files beat ten mediocre ones — stop when context is good enough.
