skillsSource-backedReview first Safety ✓ Privacy ✓
Prompt Caching Cost Audit Capability Pack Skill
Expert prompt caching cost audit capability pack for measuring Claude Code cache hit rates, identifying cache invalidation triggers, and reducing token spend from CLAUDE.md edits, model switches, and startup load changes.
HarnessClaude CodeCodexWindsurfGeminiCursorCLI
Level:expertType:capability-packVerified:validated
Review first — review before installing
Open the source and read safety notes before installing.
Safety notes
- Cost reduction must not remove safety-critical instructions from CLAUDE.md or skills without explicit review.
- Aggressive cache optimization that strips needed context can increase mistake rates and rework cost.
- Model switches invalidate prompt caches; frequent alias changes can silently increase spend.
- This skill recommends configuration changes; it must not edit CLAUDE.md, skills, or MCP config without explicit user approval.
Privacy notes
- Cost audits may expose repository names, team usage patterns, and internal project structure from CLAUDE.md and skill inventories.
- Usage dashboards and `/cost` output can include account identifiers that should not be pasted into public issues.
- Public audit summaries should describe invalidation categories and recommended actions, not full settings dumps.
Prerequisites
- An active or recent Claude Code session with observed cost increase or low cache reuse.
- Access to cost or usage indicators such as `/cost`, billing dashboards, or session token summaries when available.
- Redacted view of CLAUDE.md, skills, MCP config, and settings changed recently in the workspace.
- Knowledge of model alias or version switches during the affected session window.
Schema details
- Install type
- package
- Reading time
- 9 min
- Difficulty score
- 80
- Troubleshooting
- Yes
- Breaking changes
- No
Source repository stats
- Scope
- Source repo
Package metadata
Skill and platform metadata
- Skill type
- capability-pack
- Skill level
- expert
- Verification
- validated
- Verified at
- 2026-06-14
Retrieval sources
https://code.claude.com/docs/en/prompt-cachinghttps://code.claude.com/docs/en/costshttps://code.claude.com/docs/en/skillshttps://code.claude.com/docs/en/features-overviewhttps://github.com/anthropics/claude-codehttps://developers.google.com/search/docs/fundamentals/creating-helpful-content
Tested platforms
ClaudeClaude CodeCodexCursorWindsurfGeneric AGENTS
| Platform | Support | Install path |
|---|---|---|
| claude-code | Native | .claude/skills/<skill-name>/SKILL.md |
| codex | Native | .agents/skills/<skill-name>/SKILL.md |
| windsurf | Native | .windsurf/skills/<skill-name>/SKILL.md |
| gemini | Native | .gemini/skills/<skill-name>/SKILL.md or .agents/skills/<skill-name>/SKILL.md |
| cursor | Adapter | .cursor/rules/<skill-name>.mdc |
| cli | Manual | AGENTS.md or tool-specific context file |
Full copyable content
# Trigger
"Apply the prompt caching cost audit capability pack to this session."
# Required output
1) Cache efficiency symptom summary
2) Invalidation trigger inventory
3) Startup load and settings change review
4) Minimal cost reduction plan
5) Privacy-safe audit summaryAbout this resource
Knowledge Freshness
This capability pack is grounded in Claude Code prompt caching, costs, skills, and features overview documentation verified on 2026-06-14. Cache pricing and invalidation rules can change with model releases; prefer live official docs over remembered cache TTL assumptions.
Retrieval Sources
- https://code.claude.com/docs/en/prompt-caching
- https://code.claude.com/docs/en/costs
- https://code.claude.com/docs/en/skills
- https://code.claude.com/docs/en/features-overview
- https://github.com/anthropics/claude-code
- https://developers.google.com/search/docs/fundamentals/creating-helpful-content
Source Verification Notes
Verified against official Claude Code prompt caching and cost documentation and
the public Anthropic claude-code repository on 2026-06-14:
- Claude Code prompt caching reuses stable prefix content across turns when cache keys remain valid, reducing input token cost for repeated system and project context.
- Changes to model selection, large CLAUDE.md edits, settings updates, and startup load modifications can invalidate cached prefixes and cause expensive uncached turns.
- CLAUDE.md, skill descriptions, and MCP tool schemas contribute to startup prefix content loaded on many requests when configured for automatic loading.
- Official cost documentation describes monitoring spend and factors that affect token usage in Claude Code sessions.
- Prompt caching benefits depend on stable, reasonably sized prefix content; volatile or oversized startup load reduces effective cache efficiency.
Scope Note
This complements context window audit skills by focusing specifically on cache invalidation and cost drivers rather than general context pressure management.
Core Workflow
- Capture cost symptoms: sudden spend spikes, poor cache reuse indicators, or rising input tokens after configuration changes.
- Establish baseline: note model alias, session age, recent
/costor billing observations, and timeframe of the increase. - Inventory invalidation triggers: model switches, CLAUDE.md edits, skill additions, MCP server changes, hook output changes, and settings updates.
- Review startup load stability: count skills with model invocation, MCP schema size, and CLAUDE.md length contributing to the cached prefix.
- Check session behavior: frequent unrelated task switches,
/clearusage patterns, and compaction events that reshape effective prefix stability. - Prioritize fixes with highest cache impact and lowest workflow risk: stabilize model alias, trim oversized descriptions, defer nonessential MCP servers, and batch CLAUDE.md edits outside active sessions.
- Recommend verification: compare cost or token metrics across a short before/after window on the same task type.
- Document residual uncacheable content that should remain for safety or quality.
- Produce a privacy-safe cost audit summary for the team.
Capability Scope
- Cache efficiency symptom analysis.
- Invalidation trigger inventory.
- Startup load and settings change review.
- Minimal cost reduction recommendations.
- Verification checklist for cache improvements.
- Privacy-safe cost audit reporting.
Compatibility
Native
- Claude Code / Claude: use as an Agent Skill when investigating session cost spikes, cache misses, or prefix instability after config changes.
Manual Adaptation
- Codex, Cursor, Windsurf, and Generic AGENTS workflows: use the workflow as a deterministic prompt caching audit checklist in platform runbooks.
Required Inputs
- Cost or token usage observations for the affected period.
- List of recent CLAUDE.md, skill, MCP, model, and settings changes.
- Startup load inventory when available from
/contextor equivalent. - Task type used for before/after verification.
Production Rules
- Do not sacrifice safety-critical instructions solely to improve cache hit rate.
- Stabilize model alias during long task slices to avoid unnecessary invalidation.
- Batch CLAUDE.md and skill changes outside active high-volume sessions when possible.
- Remove unused MCP servers and disable model invocation for rarely used skills.
- Treat cost metrics as directional; confirm improvements on repeated similar tasks.
- Redact account and billing identifiers from public summaries.
- Re-audit after major Claude Code upgrades or model roster changes.
Review Matrix
| Trigger | Cache impact | Safer mitigation |
|---|---|---|
| Model switch mid-session | High | Hold alias stable for task slice |
| Large CLAUDE.md edit | High | Batch edits; restart session cleanly |
| New MCP server | Medium–High | Add only when needed; remove unused |
| Skill description bloat | Medium | Trim descriptions; user-only flag |
| Frequent unrelated tasks | Medium | Use /clear between task types |
| Hook output injection | Medium | Reduce verbose hook returns |
Output Contract
- Cache efficiency symptom summary.
- Invalidation trigger inventory ranked by impact.
- Startup load and settings change review.
- Minimal cost reduction plan with rollback notes.
- Verification checklist for next session window.
- Privacy-safe audit summary.
Duplicate Check
Checked content/skills, content/guides, generated catalog text, and open
pull requests for prompt caching cost audit, cache invalidation review, and
Claude Code /cost workflows. Context window audit skills cover broader pressure
management; no skills entry provides a dedicated prompt caching cost audit
capability pack with invalidation matrix and output contract.
Editorial Disclosure
Submitted as an independent source-backed HeyClaude content entry by
kiannidev. It is based on public Claude Code documentation, the public
Anthropic claude-code repository, and Google Search Central helpful-content
guidance. No paid placement, referral link, affiliate link, or vendor
sponsorship is used.
Source citations
Add this badge to your README
Show that Prompt Caching Cost Audit Capability Pack Skill is listed on HeyClaude. Paste this Markdown into your README — it renders the badge and links back to this page.
[](https://heyclau.de/entry/skills/prompt-caching-cost-audit-capability-pack)Signals
Loading live community signals…
More like this, weekly
A short, calm digest of reviewed Claude resources. Unsubscribe any time.