negframe-score
Skill49 starsChanged 14 days agoApache-2.0View on GitHub
One repeatable pass that keeps agent-steer prose leading with the AFFORDANCE, not the prohibition. Runs `fak score negframe` over the steer-prose corpus (AGENTS.md, CLAUDE.md, the skills, or explicit paths), reads the negframe_debt (mechanical negatives with an unambiguous positive rewrite) plus the judgement-tier soft signal, retires the mechanical debt worst-first by applying the suggested reframe, checks the `--since <ref>` ratchet before landing a steer-prose change, re-measures to PROVE the debt dropped, and commits only the scorecard lane by explicit path. Use. 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
- dojo-rsi-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
- 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
5 rules. About 0 are common, 2 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
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
- Distinctive: line 31
Every finding lands in exactly one tier - do not conflate them:
- Very specific: line 56
Default scope is the steer-prose corpus (negframe.ResolveTargets: AGENTS.md, CLAUDE.md, the skills).
- Distinctive: line 67
Leave a genuinely load-bearing prohibition alone if reframing it would blur the boundary it exists to draw - this tier is a nudge, not a …
- Very specific: line 119
After editing AGENTS.md, CLAUDE.md, or a skill's SKILL.md - run --since before landing to make sure the change did not introduce a new …
- Very specific: line 121
After adding a new skill or a new refusal/guard message - scope negframe at the new file directly (go run ./cmd/fak score negframe …
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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
/negframe-score - lead with the affordance, not the prohibition
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 scorecard exists
Prose that steers an agent lands better when it leads with the ACTION to take instead of
the thing to avoid. A reminder framed as an omission to fix ("remember: the commit stamp
often gets skipped") makes the reader build a negative and invert it; "remember to stamp
the commit" is the instruction already in the form the reader acts on. Across a whole
steer-prose corpus - the guard runtime prose injected into
every session, AGENTS.md, CLAUDE.md, the skills, the refusal vocabulary - that inversion
tax compounds. internal/negframe is a TEXT-READING card: the prose on disk IS the data,
so a finding can only be retired by reframing the sentence, never by editing a JSON file.
The two tiers (read this before touching anything)
Every finding lands in exactly one tier - do not conflate them:
- MECHANICAL (HARD, gates
negframe_debt) - the finding matches a fixed idiom with an unambiguous positive rewrite (a "forget-to-remember-to" idiom, or a double negative like "not un-<adjective>" that collapses to the plain adjective). The card hands back the exactSuggeststring; apply it verbatim or a close paraphrase that preserves the meaning. This is the only tier that counts as debt. - JUDGEMENT (SOFT, advisory only) - negatively framed (a bare "never merge without
review", "not allowed") but with no confident mechanical rewrite. The card names the
Category(prohibition / absence / refusal / hedge) and a one-lineHint- the positive SHAPE to aim for - but never a literal reframe, and it never gates. Fix the cheap real ones by hand; do not chase this tier to zero, and never invent a mechanical rule just to move a soft count (see anti-gaming below).
The pass (the shared five-step loop)
Run it - from the repo root:
go run ./cmd/fak score negframe # human scorecard: debt + per-KPI go run ./cmd/fak score negframe --json # machine payload (Corpus.negframe_debt / grade) go run ./cmd/fak score negframe --per-doc # per-document finding breakdown (mechanical + judgement) go run ./cmd/fak score negframe --suggest # review-diff view: every mechanical finding with its reframeDefault scope is the steer-prose corpus (
negframe.ResolveTargets: AGENTS.md, CLAUDE.md, the skills); pass explicit paths as positional args to score anything else.--per-docalso takes--jsonfor a machine-readable finding list;--suggestprints text only.- Retire
negframe_debtworst-first - for every mechanical finding--suggestlists, apply the printed reframe to the source prose. This pass changes ONLY the flagged clause's wording - never the meaning, a claim, a number, or a link the sentence carries. - Weigh the judgement (SOFT) findings, then stop -
--per-docshows each judgement finding'sHint. Reframe the cheap, obviously-better ones by hand (state the permitted path before the boundary; assert a property directly instead of by double negative). Leave a genuinely load-bearing prohibition alone if reframing it would blur the boundary it exists to draw - this tier is a nudge, not a work-list to drain. Check the diff-scoped ratchet before landing a steer-prose change:
go run ./cmd/fak score negframe --since HEAD # or the branch point / PR base refThis reports only negatives your CHANGE introduced (exit 1 if any new mechanical one appears) - a file absent at
<ref>is treated as wholly new. This is the CI gate shape; run it locally before committing so you never introduce a new mechanical negative in the same diff that is supposed to be reducing them.Re-measure + prove, then commit only the scorecard lane, by explicit path:
go run ./cmd/fak score negframe --json > /tmp/after.json go run ./cmd/fak score negframe --compare /tmp/before.json # prints the debt delta go run ./cmd/fak score negframe --markdown > docs/NEGFRAME-SCORECARD.md fak sync reconcile --apply fak commit --path <reframe-files> --path docs/NEGFRAME-SCORECARD.md -m "style(negframe): reframe prose (fak negframe)" fak sync pushState the delta plainly:
negframe_debt N -> N' (-k). Commit the prose files you reframed together with the regenerated snapshot (if any); nevergit add -A. Push viafak sync push. End the subject with a(fak negframe)(or the touched leaf's) trailer sodos commit-auditbinds it.
Reading the payload
Corpus carries negframe_debt (the mechanical count - the debt), documents (how many
files were scanned), mechanical_debt, and judgement_soft. Schema is
fak-negframe-scorecard/1. DebtKey is "negframe_debt" - the same string
internal/scorecardpane folds into the portfolio total under the negframe card.
The anti-gaming rule (specific to this surface)
Fix a negative by reframing the sentence to genuinely lead with the action, never by
deleting the directive or muting the detector. A mechanical finding is retired by applying
its Suggest (or an equivalent positive rewrite), not by rewording it into a shape the
lexicon happens not to match. Never widen a mechanical reframeRule's pattern just to make a
finding disappear - the existing rules are deliberately high-precision (see the not un-
allowlist in internal/negframe/negframe.go) precisely so a mechanical rewrite can never be
wrong; loosening one to chase debt risks emitting a garbled "reframe" on prose it should not
have matched. If a genuinely equivalent phrasing is already positive but still flags, the fix
is to teach the lexicon that shape - recognizing real positive framing the rules were blind
to is not weakening the check; suppressing a real negative is.
When to use this
- After editing AGENTS.md, CLAUDE.md, or a skill's SKILL.md - run
--sincebefore landing to make sure the change did not introduce a new mechanical negative. - After adding a new skill or a new refusal/guard message - scope negframe at the new file
directly (
go run ./cmd/fak score negframe path/to/new/doc.md) before it joins the corpus. - On a
/loopcadence to keep negframe_debt at zero as the steer-prose corpus grows.
The scorecard is read-only; this skill's only writes are the reframed prose, the regenerated
docs/NEGFRAME-SCORECARD.md snapshot, and internal/negframe itself when a real lexicon gap
is found (never a loosened one).
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.

