drillcard: the security workflow
job
run a repeatable sequence of drills on the user's own repo so a security pass always covers the same ground and ends in a fix plus proof, not a pile of output.
gap it targets
- evidence (public record): many agent security skill packs only help when heavy extra binaries are already installed, and loading dozens of skills at once dilutes them. drillcard has six drills and each works with grep, file reads, and whatever is installed.
- these drills are written from scratch and released under MIT.
inputs
- a repo path, and a one-line statement of what the app does and who uses it.
method: the six drills
- scope-check. write down: repo, branch, languages, what is in scope, what is out. any live target must be in
keelsec.scope.toml (see probeplate). no file, no live drills.
- local-scan. run in this order:
grainseek, depfathom, hullscan, sinkmark, bilgecheck (if the repo has MCP configs, skills, or AGENTS.md). save each output as JSON.
- triage. merge by
dedup_key (kind plus path and line, or package plus advisory group). for each finding pick one: fix now, accept with reason and expiry, or false positive with a one-line reason. nothing stays unlabeled.
- fix-and-rescan. fix the "fix now" items one at a time. after each fix, rerun only the skill that found it and confirm the finding is gone and nothing new appeared.
- pr-note. write a short note: what was scanned, what was fixed, what was accepted and why, what was not covered.
- self-check. run
bilgecheck on this skill folder and the repo's agent files, so the drills themselves are not a way in.
what counts as a finding
- whatever the called skills report. drillcard adds one of its own:
ks.workflow.uncovered for any language, folder, or file type no drill covered.
output
a merged JSON array in the shared schema plus the PR note.
{"command":"drillcard","rule_id":"ks.workflow.uncovered","kind":"sast","severity":"info","confidence":"high","title":"Go files not covered by any drill","location":{"path":"services/billing/"},"sources":["drillcard"],"status":"open"}
today vs later
- today: follow the drills by hand with the sibling skills.
- once the keelsec engine exists:
keelsec scan <path> does drill 2 and the merge in one command.
stop and scope
- never skip drill 1. never mark a finding accepted without the user's reason.
- stop and ask before any fix that changes auth, payments, or data deletion logic.
1---2name: drillcard3description: "Use when the user wants a full security workflow on their own repo rather than one scan: confirm scope, run the local checks, triage, fix, re-scan, and write the PR note. It is the playbook layer that drives the other keelsec skills in order."4metadata:5 source: https://skillmd.com/skills/codejunkie99/drillcard6---78# drillcard: the security workflow910## job11run a repeatable sequence of drills on the user's own repo so a security pass always covers the same ground and ends in a fix plus proof, not a pile of output.1213## gap it targets14- evidence (public record): many agent security skill packs only help when heavy extra binaries are already installed, and loading dozens of skills at once dilutes them. drillcard has six drills and each works with grep, file reads, and whatever is installed.15- these drills are written from scratch and released under MIT.1617## inputs18- a repo path, and a one-line statement of what the app does and who uses it.1920## method: the six drills211. **scope-check.** write down: repo, branch, languages, what is in scope, what is out. any live target must be in `keelsec.scope.toml` (see probeplate). no file, no live drills.222. **local-scan.** run in this order: `grainseek`, `depfathom`, `hullscan`, `sinkmark`, `bilgecheck` (if the repo has MCP configs, skills, or AGENTS.md). save each output as JSON.233. **triage.** merge by `dedup_key` (kind plus path and line, or package plus advisory group). for each finding pick one: fix now, accept with reason and expiry, or false positive with a one-line reason. nothing stays unlabeled.244. **fix-and-rescan.** fix the "fix now" items one at a time. after each fix, rerun only the skill that found it and confirm the finding is gone and nothing new appeared.255. **pr-note.** write a short note: what was scanned, what was fixed, what was accepted and why, what was not covered.266. **self-check.** run `bilgecheck` on this skill folder and the repo's agent files, so the drills themselves are not a way in.2728## what counts as a finding29- whatever the called skills report. drillcard adds one of its own: `ks.workflow.uncovered` for any language, folder, or file type no drill covered.3031## output32a merged JSON array in the shared schema plus the PR note.33```json34{"command":"drillcard","rule_id":"ks.workflow.uncovered","kind":"sast","severity":"info","confidence":"high","title":"Go files not covered by any drill","location":{"path":"services/billing/"},"sources":["drillcard"],"status":"open"}35```3637## today vs later38- today: follow the drills by hand with the sibling skills.39- once the keelsec engine exists: `keelsec scan <path>` does drill 2 and the merge in one command.4041## stop and scope42- never skip drill 1. never mark a finding accepted without the user's reason.43- stop and ask before any fix that changes auth, payments, or data deletion logic.