Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add ychampion/cskill-agents --skill largest-first-context-sheddinggit clone --depth 1 https://github.com/ychampion/cskill-agentsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/ychampion/cskill-agents/largest-first-context-shedding)<a href="https://agentmods.dev/skills/ychampion/cskill-agents/largest-first-context-shedding"><img src="https://agentmods.dev/badge/skills/ychampion/cskill-agents/largest-first-context-shedding/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/ychampion/cskill-agents/largest-first-context-shedding"><img src="https://agentmods.dev/badge/skills/ychampion/cskill-agents/largest-first-context-shedding.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00043 | $0.00462 |
| Opus 5 | $0.00022 | $0.00231 |
| Sonnet 5 | $0.00009 | $0.00092 |
| Haiku 4.5 | $0.00004 | $0.00046 |
Grade A, and why
largest-first-context-shedding scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 6d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 30 lines — stays where its author put it; the contents beside it link to each section on GitHub.
SKILL: Largest First Context Shedding
Domain: tool-result-budget Trigger: Apply when one outbound user message exceeds the tool-result budget and you must decide which fresh results to replace first. Source Pattern: Distilled from reviewed per-message budgeting and tool-result persistence implementations.
Core Method
Gather eligible tool-result candidates per API-level message, keep track of the frozen size contributed by already-seen results, and sort the fresh candidates in descending size order. Subtract each candidate from the running total until the total drops beneath the budget. Replace only the selected large entries with persisted preview messages while leaving smaller or previously seen blocks untouched so the largest tokens are shed first.
Key Rules
- Operate on the fresh subset so replacements never revisit previously frozen decisions.
- Sort descending by reported
sizeand keep removing the largest entries, approximating their savings by subtracting their full size before the actual persistence preview is known. - Always consider the frozen size when checking the over-budget condition; if the frozen entries alone exceed the limit, accept the overage rather than persisting already-broadcast content.
- Stop selecting once the running total is under the limit so you do not over-prune and unnecessarily persist more results than needed.
Example Application
A user turn emits four tool-result blocks totaling 120k characters while the per-message budget is 50k. Select the two largest fresh results for persistence first, replace them with previews, and keep the smaller blocks inline if they still fit.
Anti-Patterns (What NOT to do)
- Do not persist the smallest results first; that keeps the biggest token consumers in the context and never brings the message within budget.
- Do not recompute the sort on the entire conversation; the budget applies per message and frozen entries should not be reshuffled.
- Do not forget to subtract the frozen size when deciding whether to select more candidates; otherwise the budget loop will persist more content than necessary because it ignores what is already over the limit.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 6d ago First seen · 30 lines · 43 tokens per session scan A c619c182d203
largest-first-context-shedding is a skill published in the GitHub repository ychampion/cskill-agents (36 stars, last pushed 5mo ago), licensed MIT. It adds 43 tokens to every session and 462 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.
Other skills, from other repositories
grape
Use Grape MCP for Codex context continuity in coding repositories. Use when a task needs repeated-turn context, omitted context restore, stale-context checks, invalidation checks, or safe continuity across branch and dirty-worktree changes.
review
Validate plans, execution, or PRs against wish criteria — returns SHIP / FIX-FIRST / BLOCKED with severity-tagged gaps.
work
Execute an approved wish plan — orchestrate subagents per task group with fix loops, validation, and review handoff.
brainstorm
Explore ambiguous or early-stage ideas interactively — tracks wish-readiness and crystallizes into a design for wish.
genie
Entry point for Genie operations — routes bug reports, questions, and operational commands, resumes existing lifecycle state, and orchestrates work that needs durable planning or coordination. Other ordinary requests bypass the lifecycle with a one-line notice unless the user asks for Genie.
wish
Convert an idea into a structured wish plan with scope, acceptance criteria, and execution groups for work.