agentleFS
odinfreeodinfree/storm-ecdsa-harness/skills/leader-rebase-stack/SKILL.md

leader-rebase-stack

Skill4 starsChanged 21 days agoMITView on GitHub

Maintain your contributions to a moving-leader leaderboard as an ordered stack of idempotent patches (stack/NN-slug/) that rebase onto whatever submission currently leads, with a background leader watcher, a restack that scores the leader first, and a submission gate tied to the live board (submit only when the stack beats the leader by a written margin at submission time). Use on any benchmark where rivals keep landing rows while you build.

The file

Files
SKILL.md

Huckepack (leader-rebase-stack)

Distilled from jackylee0424's public shor-ecdlp-contest lab (MIT, commit 15db543): his patches survive leader changes because they were never attached to a checkout — only to the idea of the current leader.

Each contribution is a patch, not a checkout

stack/<NN>-<slug>/
  apply.sh    # idempotent; edits only the bench's editable paths; no prior state
  check.sh    # optional fast assertion that the patch is present after apply
  src/        # sources/templates the apply script installs into tracked files
  PATCH.md    # what it changes, expected Δ, proof anchors, owner brief, last leader it applied to
  ENABLED     # present = included in restacks; delete to park
  • apply.sh exit codes: 0 = applied, 3 = absorbed/superseded by the leader (skip), anything else = broken (stop the restack). When a patch exits 3, update its PATCH.md and the lever ledger — the leader ate your lever, which is also a win.
  • Package only into files the leader already tracks; new untracked files fail source gates that scan the whole filesystem.
  • One builder agent per checkout (bench, bench-m, bench-c…) so parallel work never fights over one working tree.

Restack on every leader change

  1. A background watcher polls the board and exits on change (watch-leader.sh <known-best> every ~15 min; on change it prints the new best and its source commit).
  2. restack.sh <leader-commit>: reset a dedicated restack checkout to the leader, build and score the untouched leader first (you need its real number, not its README number), then apply each ENABLED patch in NN order, run the boundary and axiom gates, score, and write a timestamped report with the % cut.
  3. Parked and absorbed patches are recorded in the report, not silently dropped.

The submission gate is tied to the live board

Submit only when the stacked score beats the current leader by a written margin (his: >10%) measured at submission time — not the margin you had when the build started. A board that moved while you built is a reason to restack, not to submit.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.