keyprove: classify and confirm leaked credentials
job
turn a list of found secrets into an ordered rotation list: what it is, what it can reach, whether it is still live, and what to do first.
gap it targets
- evidence (public record): the common open tool that checks key liveness is AGPL, which is hard to embed in a hosted product. keyprove is MIT and offline by default.
- hypothesis: ranking by blast radius and liveness gets keys rotated sooner than a flat list. not measured.
inputs
- grainseek JSON output, or a list of findings with path, line, and a redacted preview.
- for live checks: a written yes from the user that names each provider.
method
- classify each finding by shape and nearby names: provider, key type (access key, token, private key, webhook URL), likely scope (read, write, admin, unknown).
- score blast radius: cloud root or admin keys and payment keys first, then source control tokens with write scope, then chat webhooks, then read-only keys.
- note exposure: in history only, in the current tree, in a public repo, in a built image.
- offline liveness hints: is there a later commit that removes or rotates it, is it
example: true, is the value a known placeholder.
- live check, only if the user said yes for that provider: one read-only identity call, for example AWS
sts get-caller-identity or a GitHub GET /user with the token. no other call. record only "live" or "not live" and the account ID with all but the last 4 characters masked.
- write a rotation list: provider, where to revoke, what to update after rotating.
what counts as a finding
- each credential, with a priority. a confirmed-live key is
critical regardless of type.
output
{"command":"keyprove","rule_id":"ks.secret.live-check","kind":"secret","severity":"critical","confidence":"high","location":{"path":".env.production","start_line":3},"evidence":"aws access key, live, account ****1234","status":"confirmed","sources":["grainseek","keyprove"]}
today vs later
- today: this skill on top of grainseek output, with the provider CLIs the user already has.
- once the keelsec engine exists:
keelsec keyprove findings.json --verify aws,github with per-provider opt-in from the scope file.
stop and scope
- no live check without a per-provider yes. never use a found key for anything except the one identity call. never paste a full key into chat, logs, or files.
- keys that look like they belong to someone else (a vendor's key in a sample) are reported, not tested.
1---2name: keyprove3description: "Use after a secret scan, to sort each found credential by provider and type, judge how bad a leak would be, and, only with the user's explicit yes per provider, check whether a key is still live using a read-only identity call."4metadata:5 source: https://skillmd.com/skills/codejunkie99/keyprove6---78# keyprove: classify and confirm leaked credentials910## job11turn a list of found secrets into an ordered rotation list: what it is, what it can reach, whether it is still live, and what to do first.1213## gap it targets14- evidence (public record): the common open tool that checks key liveness is AGPL, which is hard to embed in a hosted product. keyprove is MIT and offline by default.15- hypothesis: ranking by blast radius and liveness gets keys rotated sooner than a flat list. not measured.1617## inputs18- grainseek JSON output, or a list of findings with path, line, and a redacted preview.19- for live checks: a written yes from the user that names each provider.2021## method221. classify each finding by shape and nearby names: provider, key type (access key, token, private key, webhook URL), likely scope (read, write, admin, unknown).232. score blast radius: cloud root or admin keys and payment keys first, then source control tokens with write scope, then chat webhooks, then read-only keys.243. note exposure: in history only, in the current tree, in a public repo, in a built image.254. offline liveness hints: is there a later commit that removes or rotates it, is it `example: true`, is the value a known placeholder.265. live check, only if the user said yes for that provider: one read-only identity call, for example AWS `sts get-caller-identity` or a GitHub `GET /user` with the token. no other call. record only "live" or "not live" and the account ID with all but the last 4 characters masked.276. write a rotation list: provider, where to revoke, what to update after rotating.2829## what counts as a finding30- each credential, with a priority. a confirmed-live key is `critical` regardless of type.3132## output33```json34{"command":"keyprove","rule_id":"ks.secret.live-check","kind":"secret","severity":"critical","confidence":"high","location":{"path":".env.production","start_line":3},"evidence":"aws access key, live, account ****1234","status":"confirmed","sources":["grainseek","keyprove"]}35```3637## today vs later38- today: this skill on top of grainseek output, with the provider CLIs the user already has.39- once the keelsec engine exists: `keelsec keyprove findings.json --verify aws,github` with per-provider opt-in from the scope file.4041## stop and scope42- no live check without a per-provider yes. never use a found key for anything except the one identity call. never paste a full key into chat, logs, or files.43- keys that look like they belong to someone else (a vendor's key in a sample) are reported, not tested.