agentleFS
tuana-vntuana-vn/system-reverse-engineer/.github/skills/claim-verification-and-promotion/SKILL.md

claim-verification-and-promotion

Skill0 starsChanged 2 months agoMITView on GitHub

Verify candidate reverse-engineering claims against current source, actively search for counterexamples, narrow scope, and promote only defensible claims into canonical baseline memory.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 5 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not transfer a condition from one surface to another without evidence.

    line 73
  2. Distinctive:

    do not require an explicit output-layer condition if upstream flow already guarantees the result

    line 102
  3. Distinctive:

    if a material provenance link is missing, do not PROMOTE

    line 104
  4. Distinctive:

    Only PROMOTED claims may be copied into:

    line 135
  5. Very specific:

    If an applicable item is unresolved, use VERIFIED_NOT_PROMOTED, KEEP AS HYPOTHESIS, or another non-promoted state as appropriate.

    line 157
  6. Distinctive:

    When a promoted/candidate claim is challenged:

    line 164

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Claim Verification and Promotion — 4.0

Mission

Take candidate claims from:

  • 00_hypotheses.md
  • detailed reverse-engineering docs
  • pending evidence rows

and determine whether they deserve canonical promotion.

For Each Claim

1. Restate Precisely

Avoid words like:

  • always
  • all
  • never
  • global
  • only

unless proven.

2. Supporting Evidence

Trace exact current source:

  • caller
  • condition
  • selector
  • implementation
  • boundary
  • config as relevant.

Material source-backed claims must record a repo-relative source path. When a specific class/method/function/config key is involved, record the symbol/key as well. Line/range is preferred when practical.

Evidence precision hierarchy:

path + symbol + line/range   preferred
path + symbol                strong
path                         minimum for material source-backed claim
class name only              insufficient when path is discoverable

3. Counterexample Search

Actively search:

  • other implementations
  • other call sites
  • alternate branches
  • config-driven paths
  • target/model/version/mode variations
  • fallback/retry
  • reflection/service loader
  • tests hinting at alternate production behavior

4. Scope Check

Narrow the claim to the actual proven domain.

For claims spanning multiple endpoints/operations/modes, create a compact scope matrix:

Surface / OperationClaim Applies?ConditionEvidence

Do not transfer a condition from one surface to another without evidence.

4A. Field / Data Provenance Gate

Mandatory for claims about:

  • response/output fields
  • conditionally exposed data
  • persisted/configured values
  • messages/events
  • identifiers/normalized values
  • downstream-derived state

Prove:

trigger / request condition
→ source/downstream retrieval
→ source field/data availability
→ parse/decode
→ transform/normalize
→ mapping/storage
→ serialization/output
→ observable result

Rules:

  • conditional output may be implemented indirectly by conditional retrieval
  • do not require an explicit output-layer condition if upstream flow already guarantees the result
  • implementation shape alone is not enough to promote an observable-behavior claim
  • if a material provenance link is missing, do not PROMOTE

5. Baseline Version

Record current commit/branch if available.

6. Decision

PROMOTE

Requirements satisfied, including provenance when applicable.

VERIFIED_NOT_PROMOTED

Evidence supports the claim but one promotion governance requirement is intentionally deferred.

KEEP AS HYPOTHESIS

Useful but incomplete.

REJECT

Contradicted or unsupported.

Update Ledger

Record:

  • verification method
  • counterexample search
  • result
  • status
  • supersession if applicable

Canonical Promotion

Only PROMOTED claims may be copied into:

  • 00_current_understanding.md
  • 00_master_decision_matrix.md

Canonical entries must reference Evidence ID.

Promotion Completion Gate

Before PROMOTE:

[ ] claim wording is precise and scope-limited
[ ] current-source support exists
[ ] counterexamples/alternate paths were searched
[ ] scope conditions were not borrowed from another surface
[ ] runtime wiring is proven where relevant
[ ] field/data provenance is proven for observable-data claims
[ ] no material evidence link remains unresolved
[ ] material source evidence includes repo-relative path
[ ] symbol/config key is recorded where applicable

If an applicable item is unresolved, use VERIFIED_NOT_PROMOTED, KEEP AS HYPOTHESIS, or another non-promoted state as appropriate.

Challenge Reverification

A user correction or reviewer disagreement is not evidence.

When a promoted/candidate claim is challenged:

  • reopen the claim
  • re-trace current source
  • search counterevidence
  • revise only if evidence supports revision

Important

A claim becoming familiar through repetition is not evidence.

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.