Cilytics. CI health across flisk-sh, every hour.
Every active public repository in the flisk-sh org, read from the GitHub API once an hour: what is red on the default branch right now, which workflows flip between pass and fail, which are slow, and Flisk's own deterministic scoreboard. No model, no account.
Trend
One point per hourly snapshot, whole org.
Red now
Workflows whose newest completed run on the default branch failed or timed out. The strip is the last 20 runs, oldest left.
Per repo
Runs per day for the last 14 days (green passed, red failed), pass rate of the last 100 runs and median run time. Red repos first.
Flaky
Workflows that went from pass to fail or back at least three times in their last 100 runs, with no change in between that we can see. Most flips on top.
Slow
The slowest workflows by median run time. The dashed outline is the 90th percentile run.
Outcomes
Runs per day on default branches, and how every run in the window ended.
AI-SDLC
Is each repo ready for coding agents, and do Flisk's deterministic checks hold the harness together? Instructions, specs, tools, guardrails and verification, after Anthropic's AI-SDLC guidance. Repo flags are read from each repo's root folder; scorers come from flisk-sh/flisk.
Scoreboard
Flisk's deterministic scorers from scripts/scores/run.mjs in flisk-sh/flisk, the same set every PR runs. Gates and ratchets fail CI; reports only inform.