[Bug] Skill listing truncated with descriptions dropped due to context budget constraints
Status Fixed / completed
Reported on v2.1.153
Maintainer reply ✓ Yes — bcherny
Activity 3 comments · opened Jun 13, 2026 · closed Aug 19, 2026
💡 Likely answer: A maintainer (bcherny, collaborator)
responded on this thread — see the highlighted reply below.
Bug Description
Skill listing will be truncated
81 descriptions dropped (full descriptions kept for most-used skills) (3.2%/1% of
context): fixbill, prototype, review, +78 more
run /skills to disable some, or raise skillListingBudgetFraction (currently 1%)
in settings.json
Opting in would cost ~6k tokens for skills every session and uses rate limits
faster
Environment Info
- Platform: darwin
- Terminal: Apple_Terminal
- Version: 2.1.153
- Feedback ID: 8f485742-ea68-4e9d-87d3-9aa89febeb59
Errors
[]
3 Comments
Adding a data point on the silent variant of this truncation (no warning shown at all, unlike the message quoted in this issue).
Setup: repo with ~50 project-level skills under
.claude/skills/, no plugins. A local script sumsdescription:field lengths across allSKILL.mdfrontmatter to track the aggregate character budget.Observations across sessions on the same repo:
<system-reminder>skills listing, no truncation, no warning./doctornotice, nothing in the statusline — matches the pattern described in #68677 rather than the explicit warning shown in this issue's report.So on this install/version, the cap sits somewhere in (22,924, 23,400] cumulative description chars, and truncation is happening with zero user-facing signal — which lines up with #64606's "silently truncates ... without any warning" description of the underlying
skillListingBudgetFractionmechanism, just without the message this issue shows. Possibly two different code paths (or versions) hit the same budget differently: one warns, one doesn't.Caveats: char counts are from a repo-local script, not the harness's internal accounting, so the exact threshold value may not transfer to other setups (token-based counting, other context consumers, model/context-window size could all shift it). Not filing as a separate issue since #64606/#68677/#68135 already cover the mechanism and both truncation behaviors — just adding a reproducible bound in case it's useful for narrowing down why the warning fires in some cases and not others.
This is expected behavior: the skill listing has a context budget (default 1% of the model's context window) and least-used skill descriptions are shortened first — you can raise
skillListingBudgetFractionin settings or trim skill descriptions, as described at https://code.claude.com/docs/en/skills#skill-descriptions-are-cut-short. Closing — reply to reopen.🤖 Generated with Claude Code
This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.