deterministic profiler - by the Spec Kitty team
What's wasting your Claude Code, Claude Desktop, Codex, GitHub Copilot, OpenCode, Cursor, Kiro, or Antigravity tokens?
Run a safe local scan. Get a custom plugin built from your own coding logs.
Agent Analyzer finds the exact token leaks in your recent sessions, then turns those findings into a personalized setup pack for Claude Code, Codex, GitHub Copilot, OpenCode, Cursor, Kiro, Antigravity, and Claude Desktop MCP.
What Agent Analyzer detects
- Detect context bloat and retry loops
- Find wasted reads and oversized outputs
- Download a plugin matched to your personal token leaks
Privacy and trust guarantees
- No account required
- Raw logs stay local
- No secrets uploaded.
Your machine does the analysis; only the reviewed sanitized JSON report uploads.
start here
Run the safe local scan
npx --yes agent-analyzer@latest run
Read-only local scan: no hooks, settings changes, or repo edits. It asks before uploading only the sanitized report JSON, then uses that report to build your personalized plugin.
Install trust details
What this command does
- 1Finds supported local agent logs.
- 2Parses metrics and redacts secrets locally.
- 3Writes a reviewable report JSON.
- 4Asks before upload.
- 5Builds a custom plugin from the leaks it found.
Verify before running
Source on GitHub npm package Security model Privacy terms Manual download Benchmark proofoutcome preview
One command finds your leaks and prepares your plugin.
report preview
See the waste before changing anything.
generated plugin
Your logs become a personalized plugin.
The plugin is not a generic checklist. Agent Analyzer builds it from the token leakage in your own sanitized report, so your agent gets the skills, commands, and vetted setup guidance that match the waste patterns the scan actually found.
Findings drive the contents: repeated reads add retrieval hygiene, large shell output adds output budgeting, retry loops add a stop-and-reframe routine, and detected stacks add only matching code-intelligence or integration setup.
copy / paste
Find your leaks. Download your plugin.
npx --yes agent-analyzer@latest run
Runs on your machine first, asks before upload, sends only sanitized report JSON, and builds a plugin from your detected token leaks.
Download your custom skills and save tokens now.
custom token recovery target
-- less waste, targeted by your custom skills
Analyze locally, download the custom skills generated from your own logs, then apply the exact guards for the token leaks Agent Analyzer found.
Top Problems
Session Timeline
Estimated context/token volume by parsed turn. Green overlay = estimated tokens you can save.
generated remediation
Copy quick fixes now. Download the plugin built for your leaks.
Add the relevant AGENTS.md lines now. The generated plugin packages your own report into benchmark-backed operating guidance so future sessions spend more of your plan writing software and less on the rereads, retries, dead context, and noisy tools found in your logs.
- Session hygiene nudges.
- Scoped retrieval recommendations.
- Output-budgeted command habits.
- No unproven reducer installs.
Copy-ready AGENTS.md lines
Use only the lines that match how you work.
Recommended tools to address waste
These are not random installs. Telemetry tools are labeled as measurement only, and reducers are recommended only when our repeated runs showed savings for the matching waste signal. Use the generated plugin to turn the report into setup instructions.
No high-priority recommendation detected from current signals.
Savings validation
Environment signals
Security Receipt
portable findings
Get the report pack and your custom plugin by email
The report pack shows where your own agent sessions are leaking tokens. The generated plugin turns those specific findings into setup rules and workflow nudges using only reducers that earned scoped recommendations.
- Raw transcripts stay local.
- Email delivery includes the report pack and generated plugin as attachments.
- Benchmark methodology and primary data.