Detect stale feature flags

Overview

After you have already set cleanup expectations, stale flag detection should be handled by a reusable coding-agent skill. The agent only needs to read your cleanup rules, query FeatBit through FeatBit CLI or FeatBit MCP, search the repository for flag usage, and return a short cleanup review.

Write a stale flag review skill

By this point, the hard parts are already in place:

• Cleanup rules define when different flag types should be reviewed or removed.
• FeatBit CLI or FeatBit MCP lets the agent interact with FeatBit APIs and fetch project feature flag data.
• The remaining work is to encode your stale-flag detection logic into a small coding-agent skill, such as .agents/feature-flag/skill.md, AGENTS.md, or another project rule file.

That skill should tell the agent to:

• Fetch FeatBit flag state, including rollout state, tags, audit history, archive state, and experiment state.
• Search the repository for each flag key and related registry entry.
• Compare FeatBit state and repository usage against your cleanup expectations.
• Classify each candidate as remove, retain, or needs-owner-decision.
• Never archive a FeatBit flag until the deployed code no longer evaluates it.

The full pattern is already demonstrated in the stale flag review example, where a coding agent uses project rules, repository references, and FeatBit state to list stale feature flags.