context-compression

Context Compression

Purpose

Reduce large logs, transcripts, and outputs to a compact summary that preserves every fact needed to continue the work.

When to use it

• Long tool output, test logs, or build logs.
• Handing off context near a context-window limit.

Instructions

1. Preserve exact failures, commands, filenames, line numbers, and assertions.
2. Preserve API changes, decisions, and open risks verbatim.
3. Summarize repetitive or noisy passages; never drop a unique fact.
4. Separate observed facts, inferences, and uncertainty.
5. State explicitly what was omitted and why.

Output format

• Facts (observed, exact).
• Inferences (clearly labeled).
• Open / uncertain.
• Omitted (what was dropped and why).

Anti-patterns

• Paraphrasing an error message and losing its exact text.
• Presenting an inference as an observed fact.
• "Compressing" by deleting information that is still needed.

Verification checklist

• A new reader can identify the objective, current state, exact failures, decisions, constraints, modified files, verification results, and next action.
• Repeated material is collapsed, while every unique fact remains.
• Sensitive values are omitted or redacted without hiding that redaction occurred.

Apply context-discovery before broad scanning: use a supplied context capsule first; otherwise check context frontmatter freshness, read .agent/context/routing.md, load only relevant summaries, and verify critical claims against source.