maintain-formal-verification-skill · git:20261108.ac1df7f · 2026-11-08 · sha256 857c725d6b6821f4
maintain-formal-verification-skill git:20261108.ac1df7fA
Immutable. This exact content never changes and is served at /api/v1/blob/857c725d6b6821f4.
No tool to install first
claude plugin marketplace add https://markdownregistry.com/mp/a/art_g6qr2qmeag6totbn/857c725d6b6821f49a09c72d70b542abad4cf2445665e41dd3c5985feff55a43.json
Then /plugin install maintain-formal-verification-skill@markdownregistry-g6qr2qmeag6totbn in Claude Code. It installs this exact version, the one graded A here, pinned to the commit it was graded at, and leaves it there.
---
name: maintain-formal-verification-skill
description: "Re-sync an existing formal verification skill after the covered code changed: re-check every assumption, update cited lines, rerun the tool; formally-verified only after a green rerun."
aliases: []
metadata:
credits:
verification-maintenance: pstack
author: Lauren Tan
url: "https://github.com/cursor/plugins/tree/main/pstack"
port: pstack-claude
port-author: Michael Denyer
port-url: "https://github.com/michael-denyer/pstack-claude"
---
# Maintain Formal Verification Skill
A formal spec that cites lines that no longer exist is stale, and a stale
spec's green is worthless. This skill re-syncs an existing
`verify-formal-*` project skill (created by `create-formal-verification-skill`)
to the code it models.
## When to use
- the modeled protocol, arithmetic, or invariant changed shape
- cited file:line anchors drifted after refactors
- before trusting a green run that predates recent changes to the covered
code
## Workflow
1. **Inventory the model's bindings**: every cited file:line, every named
function, every assumption on the skill's assumption list.
2. **Re-check each binding against current code**: does the cited line still
contain the modeled step? Does the assumption still hold (pid reuse,
atomicity, integer width, timing)? A binding that no longer matches is
drift; record it, do not silently absorb it.
3. **Update the model** where drift is real: move citations, adjust steps to
the new protocol shape, and rewrite the assumption list to match reality.
The property itself does not weaken — if the new code violates the
property, that is a finding to report, not a model edit.
4. **Rerun the tool** (`TLC`/`apalache`, `lake build`, `cargo kani`) and
capture real output.
5. **Update the project skill** in place: new citations, new assumption
list, same run command — and record the resync through the ledger's
locked writer (`appendLedgerEntry` and `defaultLedgerPath` from the
installed package's `dist/features/verification-ledger/index.js`;
`babysit`'s ledger step demonstrates the exact invocation — never append
raw rows): `formally-verified` with the rerun
command as evidence only when the rerun is green; a counterexample during
resync records `verifier-failed` with the counterexample as evidence and
is reported before any model change.
## Rules
- A counterexample found during resync is reported before any model change;
verify whether the code shares the flaw.
- Never mark a resync done with an unrerun tool; the rerun is the resync.
- If more than half the bindings drifted, say so — the model may have been
modeling an implementation that no longer exists, and regeneration via
`create-formal-verification-skill` is the honest path.