01
Find token waste
usage · repeated context · retry loops
Start with visible usage signals, expensive tasks, repeated context, and retry loops so users know where tokens are going.
usage signalsexpensive taskstoken hotspots
Free AI checkup · read-only report in 3 minutes
See why your AI
gets costly and slow.
Most teams blame the model price. The quiet waste is repeated context, stale tasks, bloated tools, and invisible retry loops. Paste one prompt into the agent you already use and let SkillFM Beacon run a read-only checkup first.
The report tells users why AI slowed down, where too many tokens were spent, whether injection or abnormal process signals are visible, and when plan or credit windows need attention. Missing data is explained instead of invented.
read-only report firstno system change before consentmissing data is called out
what the checkup does
A checkup is not just a score. It does four concrete jobs.
02
Find slowdown causes
context · tools · logs
List stale worktrees, large logs, bloated prompts, and tool clutter as a read-only audit so the main agent knows why it got slow.
old worktreesprompt bloatread-only audit
03
Find safety anomalies
injection · process · port
Check visible prompt injection, abnormal processes, suspicious ports, and launch items. Missing access is explained without panic.
injection tracesabnormal processesodd ports
04
Track plan windows
plan · credits · renewal
When subscription, credit, renewal, or usage windows are visible, the report surfaces them. Missing data is explained clearly.
renewal timingcredit windowmissing data
05
Give action
today · 3 actions · next report
The report ends with the three most useful actions for today: what to save, what to stop, and what to watch next.
three actionstodayrepeatable
proof, not promise
After the prompt is pasted, users see this path.
These are redacted flow examples. The page does not show fake user counts, fake reviews, or sample output as if it came from a real customer.
Give the prompt to your current agent
No new dashboard to learn first. Use the Claude Code, Cursor, or Cline workflow already in front of you.
one prompt starts the pathActivation uses real device authorization
The agent must use the actual verification URL and code from the sidecar. Placeholder codes are forbidden.
the user completes consentPreview SAFE + Beacon first
Module install starts with dry-run so the user can see what would change before continuing.
preview before executionReturn a checkup and cleanup audit
The report lists waste, slowdown, risk, and three suggested actions. Missing data is not guessed.
the main agent can continuecopy this to your agent
Paste this into your main agent.
It starts local SkillFM Beacon. If activation is needed, it uses real device authorization. After your approval, it previews the base SAFE + Beacon module before running the read-only checkup and cleanup audit.
What happens after you paste it
- Your AI starts local Beacon
- Activation uses real device authorization
- It previews SAFE, then runs the checkup
PASTE TO AGENT
Help me connect SkillFM Beacon. Start the local sidecar first; if activation is required, use the real device authorization flow and do not invent a code. After I authorize it, call skillfm_install_module({name:"beacon", dry_run:true}) to preview the base SAFE + Beacon module, then install only after I confirm. Then run a read-only AI checkup and cleanup audit: tell me what token/usage signals are visible, whether context or tools are bloated, what safety risks exist, and the 3 most useful actions for today. Do not invent billing numbers; if data is missing, tell me what authorization is needed. Run: $ npx -y -p @skillfm/local@latest skillfm-local start
one clear pathDo one thing:
Do one thing:
paste SkillFM into the AI you already use.
01
Beacon
Run the checkup
Ask the agent: what is expensive, slow, stale, or risky today?
Health, token pressure, context load, tool load, and the single most useful fix become a report the main agent can continue from.
02
Cleanup
Audit before action
Find waste and risk before deleting files or changing setup.
Beacon lists stale tools, repeated prompts, old worktrees, and large logs. You and your main agent still decide what to change.
03
SAFE
Act only with consent
Normal chat stays light. Higher-risk actions are previewed, explained, and confirmed first.
SAFE is the boundary between SkillFM and your main agent: separate suggestion from execution so the user keeps final control.
04
Reports
Keep earning trust
If the first report is useful, open daily reports, budget alerts, and Pro automation.
SkillFM earns trust by showing visible value first, then letting users choose whether to enable deeper monitoring and stronger automation.
first report
Not another dashboard. An action report for the agent.
Beacon turns token hotspots, slowdown causes, injection or abnormal process signals, plan or credit windows, and cleanup advice into a report the main agent can continue from. The agent knows whether to save, stop, or ask for permission.
BEACONONLINE
usage watchUsage signals
Expensive tasks, repeated context, token hotspots
context watchSlowdown causes
Long logs, stale tasks, prompt bloat, tool bloat
safety watchSafety anomalies
Injection traces, abnormal processes, odd ports
01Context driftcontext driftnoticed
02Token hotspottoken hotspotroute
03Safety signalsafety signalwatch
what the first run shows
On first use, see three things right away.
Copy the prompt first and get one useful report. Reviews, user counts, and outcome claims are shown only when backed by real evidence.
01Beacon
AI Checkup
What is the agent state?See why the agent slowed down, where tokens went, and the best first fix for today.
02Beacon
Cleanup Audit
What is slowing it down?Read-only scan for stale tools, bloated prompts, broken worktrees, and large logs.
03Beacon
Safety Audit
Any injection or abnormal processes?Check visible prompt injection, abnormal processes, suspicious ports, and launch items. Missing access is called out.
04Beacon
Plan Watch
When should plan timing matter?When plan, credit, and renewal windows are visible, they are surfaced. Missing dates are not guessed.
trust boundary
Trust starts with boundaries, not slogans.
SkillFM defaults to reports and recommendations. System-changing actions such as install, cleanup, or setup must be previewed, explained, and confirmed by the user.
The first run produces a report. It does not delete files, change config, or decide for the user.
Activation, install, and cleanup follow explicit steps and stop when user confirmation is needed.
If usage, billing, or system state is not visible, the report describes the gap instead of turning guesses into numbers.
@skillfm/local is distributed through npm and GitHub Actions, so the copy command is reproducible.
who it is for
Built for people who use AI as a workbench.
If you only chat occasionally, SkillFM may be more than you need. If your agent writes code, researches, edits files, and runs workflows every day, the checkup becomes useful fast.
Good fit
Daily Claude Code / Cursor / Cline users
You want to know where cost and context load are coming from.
Multiple active projects
Old tasks, worktrees, and logs often leak into the current run.
Real workflows with AI agents
You want system actions previewed and audited before confirmation.
Not ideal yet
Light chat only
Without projects, file edits, or tools, the first checkup has less to inspect.
Expecting automatic fixes immediately
SkillFM reports and recommends first. It does not bypass user confirmation.
Looking for marketing claims
The page will not show fake reviews, fake user counts, or unverifiable outcome stories.
faq
Common Questions
Will the checkup change my system?
No. The default first run is a read-only report. Install, cleanup, or setup actions are previewed and explained before you confirm.
What if exact usage is not visible?
The report says what is visible, what is missing, and what authorization would be needed. It does not invent billing numbers.
How is this different from a monitoring dashboard?
SkillFM starts with an action report for the main agent, not another panel for the user to interpret.
Why no testimonial block yet?
Testimonials should be real USERTEST or customer quotes. For now the homepage shows redacted flow examples and a reproducible install path.
from checkup to monitoring
Run the free checkup first. Keep monitoring if it proves useful.
The first report proves whether SkillFM can find waste, slowdown, and risk in your setup. Later, users can enable daily reports, budget alerts, team rules, and Pro automation.
01Copy to AI
02Start Beacon
03Read-only checkup
04Cleanup audit
05Action advice
06Continuous monitoring