Claude/Code/Q/What does the skill listing truncation warning mean?

• What does the skill listing truncation warning in Claude/Code mean?

  • Answer

    • AI/Response
      • What the message is: At the start of a turn, Claude Code injects a skill listing into the model context: each discoverable skill’s name plus (normally) its description and when_to_use text so the model knows what exists and when to load it. When that listing would use more than your configured share of the context window, the CLI drops full descriptions for the least-recently-used skills and keeps full text only for skills it considers most-used, then prints the warning you saw (with counts and an estimate of how much of the window the full listing would have taken versus the cap).
      • Where it comes from: This behavior is implemented in Claude Code itself (not your shell). It is controlled by settings documented on Claude Code settings: skillListingBudgetFraction reserves a fraction of the model’s context window for that listing (default 0.01 = 1%). When the listing would exceed the budget, descriptions for the least-used skills are collapsed to bare names so the model can still invoke them but no longer sees why. A companion cap, maxSkillDescriptionChars (default 1536), truncates each skill’s combined description + when_to_use text before the overall budget is applied. See the same settings table rows and the skills guide Extend Claude with skills (frontmatter reference explains the listing and per-skill truncation).
      • How to read the parenthetical (example 3.2% / 1%): The first figure is an estimate of how much of the context the full skill metadata would consume; the second is your current budget (often the default 1%). The gap is why many descriptions were dropped in one shot.
      • What you can do: Raise skillListingBudgetFraction in ~/.claude/settings.json (or project or local settings) to keep more descriptions at the cost of more tokens every turn and faster consumption of rate limits—the message’s “~6k tokens” line is that tradeoff spelled out. Alternatively reduce load: shorten description / when_to_use in SKILL.md files, disable plugins or skills you do not need (the message points at /skills and the skillOverrides mechanism described in the skills doc), or raise maxSkillDescriptionChars only if individual skills are being cut short. For visibility into what was truncated, the settings doc says /doctor shows the current truncation count and which skills are affected.