agentsclimarketplace

Env driven strategy preset

Skill kjuhwa/skills-hub/skills/configuration/env-driven-strategy-preset

Self-correcting knowledge corpus for Claude Code — 9 stable shape clusters, bias-correction pipeline baked into contribution flow. 47 papers, 45 techniques, 1.1k skills.

Install
npx -y skills add kjuhwa/skills-hub --skill env-driven-strategy-preset

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Expose one env var that selects a named preset, where each preset is a tuple of behavior weights (innovate / optimize / repair). Default preset is "balanced"; operators swap presets for incident response without touching code.

SKILL.md

2.4 KB, 467 tokens by cl100k_base, as published. Nobody here has run it

Env-Driven Strategy Preset

When to use

Your loop has multiple competing intents (new features vs. hardening vs. repair). You want operators to shift the balance without a redeploy or a config file edit.

Shape

One env var (EVOLVE_STRATEGY in the source repo), finite preset names, table of weights:

PresetInnovateOptimizeRepairWhen to use
balanced (default)50%30%20%Daily operation
innovate80%15%5%System stable, ship features fast
harden20%40%40%After major changes, focus on stability
repair-only0%20%80%Emergency — all-out repair

The loop reads EVOLVE_STRATEGY once per iteration, looks up the tuple, and uses the weights as the mixing distribution for its next action selection.

Rules

  • Fail open to the default. Unknown values → balanced, logged once.
  • No partial weights from env. Operators cannot pass raw numbers; they pick a named preset. This keeps the search space small and documented.
  • Reload per iteration. The env var is re-read on each loop tick so operators can change strategy on a running daemon without restarting.
  • Presets are tuples, not booleans. The invariant innovate + optimize + repair = 100 must hold; a unit test enforces it.
  • Auto preset (optional). An auto preset can inspect recent failure rate and pick repair-only / harden / balanced dynamically, so operators don't have to flip manually during incidents.

Anti-patterns

  • Exposing each weight as its own env var. Now operators debug three knobs instead of one named dial, and the weights can go out of sync.
  • Putting the weights in code. Ops can't change them without a deploy.
  • Silent fallback on typo. Always log "unknown strategy 'X', using balanced" so misconfigs surface.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.