dojo-rsi-score
Skill49 starsChanged 14 days agoApache-2.0View on GitHub
One repeatable pass that keeps the dojo's self-improving RSI loop actually CLOSING on our own billed usage - the real calibration history the dojo measures - instead of a plan-mode scaffold that never acts. Drives the native dojo-RSI verbs (`fak dojo-rsi fold|propose|rewrite|run|loop|trend`) over a scored dojo report, picks the worst-calibrated MEASURED, NON-FLOOR cell, previews the one-literal recalibration, self-scores it on a strict FoldCalibrable drop + a sample floor + an external witness, routes REPROJECT/HARVEST/floor cells to the human/agent arm, and. Use when this named workflow matches the task.
In this repository151 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- llms-full.txtllms.txt
- llms.txtllms.txt
- agent-readinessSkill
- appeal-scoreSkill
- bottleneck-mapSkill
- ci-repairSkill
- claim-repro-scoreSkill
- clear-out-wipSkill
- commit-cleanSkill
- conflation-scoreSkill
- cross-validateSkill
- curate-clusterSkill
- debt-cleanSkill
- debt-orchestratorSkill
- disambiguation-scoreSkill
- dos-dispatch-loopSkill
- dos-dispatchSkill
- dos-next-upSkill
- dos-replanSkill
- dos-witness-claimSkill
- field-borrowSkill
- fleet-waveSkill
- git-subagent-syncSkill
- goalSkill
- guard-rsi-scoreSkill
- harness-creatorSkill
- harness-gardenSkill
- industry-scoreSkill
- issue-orchestratorSkill
- issue-queueSkill
- issue-triageSkill
- lightgap-scoreSkill
- memory-compactSkill
- milestone-scoreSkill
- modularizeSkill
- negframe-scoreSkill
- opencode-waveSkill
- operator-heaviness-scoreSkill
- persona-scoreSkill
- phased-planSkill
- plan-auditSkill
- quality-scoreSkill
- question-loopSkill
- refresh-cachedoc-numbersSkill
- refresh-readmeSkill
- releaseSkill
- resume-watchdog-auditSkill
- run-it-all-nightSkill
- score-2xSkill
- scorecardSkill
- scout-loopSkill
- signal-firstSkill
- skill-lifecycleSkill
- skill-overlapSkill
- skill-scoreSkill
- slop-scoreSkill
- sota-checkSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 0 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Very specific: line 50
RUN (writes nothing, opens no worktree, commits nothing) and refuses a floor/REPROJECT/ HARVEST cell with the reason it routes.
- Very specific: line 65
Commit only the dojo lane, by explicit path: - run fak sync reconcile --apply, then commit via fak commit --path (never git add -A), and …
- Very specific: line 86
Keep the dojo + dojocal SOURCE ASCII-only (no em-dashes / smart quotes): the repo's provenance pre-commit hook crashes on non-cp1252 UTF-8 …
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
dojo-rsi-score - the dojo's self-improving loop, scored against reality
An instance of the generic scorecard doctrine. Read that for the five laws and the RSI loop; this file is the per-surface specialization.
Why this skill exists
The dojo closes four RSI rungs - predict, run, measure, score,
trend - against the provider's billed reality. The fifth rung, act on the calibration gap
and prove the act helped, is the dojo-RSI loop. Its
twin, guard-rsi-score, keeps the guard's VERDICT loop honest; this skill keeps the dojo's
CALIBRATION loop honest.
The law, in one line: the dojo's RSI loop must close on a re-measured FoldCalibrable drop
from OUR OWN billed corpus + an external witness - never on a fabricated number, and never by
recalibrating a floor it exists to defend.
The load-bearing idea (read before touching a number)
Optimise FoldCalibrable, not raw mean_calib_err. Raw calib_err is minimised at
claimed := realized, so a constant rewrite is a free, content-free "gain". For a floor
metric (vcache-warmth/false_warm_rate claim 0.0, the lethal false-warm class; the bimodal
resume-posture/cross_session_warm_hit_rate default) closing the gap would erase the guard
the dojo exists to defend. FoldCalibrable folds a genuine estimate by its calib_err but a
floor by its breach, so a floor recalibration RAISES the fold and reverts. That is the
structural reason the loop cannot optimise itself into dishonesty - do not weaken it.
The pass (the shared loop)
- Score the corpus - produce the report the loop reads from real billed usage:
fak dojo run --corpus ~/.claude/projects --json > report.json. An empty/UNMEASURED corpus is not a failure; it is honestly uncandidatable (point the scenario at a billed corpus, never fabricate rows). - Fold the calibrable metric -
fak dojo-rsi fold --report report.jsonshows the estimate calib-err and the floor-breach term separately, so a floor breach can never hide inside an averaged-down estimate gain. - Propose worst-first -
fak dojo-rsi propose --report report.jsonranks the cells by novelty x value x staleness. Only a RECALIBRATE is self-keepable; REPROJECT / HARVEST / ROUTE_FLOOR / ROUTE_UNMEASURED are routed, never kept by the pure loop. - Preview the recalibration -
fak dojo-rsi rewrite --report report.jsonrenders the exact one-literal change tointernal/dojo/claims.goa RECALIBRATE would make. It is a DRY RUN (writes nothing, opens no worktree, commits nothing) and refuses a floor/REPROJECT/ HARVEST cell with the reason it routes. Read the diff before trusting the swap. - Self-score + check the keep-bit -
fak dojo-rsi run --report report.json --witness '{"ok":true}' --check. KEEP requires measured rows > 0 AND a strictFoldCalibrabledrop AND the per-lever sample >=--min-sample(default 3) AND the external witness.--checkre-derives those invariants from the iteration's own fields, so a fabricated KEEP is caught at the gate. Never pass a witness the loop authored. - Route what the loop could not keep (close the loop) - a ROUTE_FLOOR breach is a belief-code bug to escalate (a bug in the lever's projection, never a claim swap); a REPROJECT routes an agent patch constrained to declared paths; a HARVEST files a goal-gated issue with the board row + corpus evidence. The pure loop never auto-lands these.
- Trend + cadence stop -
fak dojo-rsi trendfolds the committeddocs/dojo/rsi-journal.jsonlinto the KEEP/REVERT/ESCALATE rollup. A freshly touched cell enters cooldown sofak dojo-rsi loopstops thrashing the same row; cooldown limits repeated action, not the unbounded discovery ledger. - Commit only the dojo lane, by explicit path - run
fak sync reconcile --apply, then commit viafak commit --path(nevergit add -A), and push viafak sync push:
The claim literal (and its pinned-claim test, when an honest recalibration updates it), the lever/registry change, the doc, this skill. End the subject withfak sync reconcile --apply fak commit --path internal/dojo/claims.go --path docs/dojo/rsi-journal.jsonl -m "fix(dojo): recalibrate <cell> (fak dojo)" fak sync push(fak dojo)or(fak dojocal).
The anti-gaming rule (specific to this surface)
Close the loop by re-pointing a genuine estimate at measured reality, never by widening the
ruler or erasing a floor. The calibration band (CalibBand / calibErr / MaxCalibErr)
lives in internal/dojo; the rewrite target is one anchored literal in claims.go. A diff
that touches the band changes a second file and fails the one-file keep-bit. Lowering a floor's
claim to its empirical rate RAISES FoldCalibrable and reverts. A KEEP requires a strict
FoldCalibrable gain on real billed bytes AND a sample floor AND a green external witness;
weakening CheckIteration to pass an unwitnessed, too-few-samples, or floor-target keep is the
one move that turns the gym into theater - exactly the auto-erasure this loop exists to forbid.
ASCII gotcha
Keep the dojo + dojocal SOURCE ASCII-only (no em-dashes / smart quotes): the repo's provenance
pre-commit hook crashes on non-cp1252 UTF-8 before its own escape fires. Use - for a dash
and straight quotes in .go / .py. (The .md docs may use UTF-8.)
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
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.

