CA Cybersecurity
Runs defensive security: alert triage, compromise investigation, attack paths, vulnerability prioritization, detection, and risk reporting. Use when an alert, a suspicious login, a reported email, encrypted files, or a possible compromise needs scoping and containment; when deciding what to patch first out of a scanner or pentest backlog; when hardening identity, endpoints, segmentation, cloud tenants, or a build pipeline; when writing or tuning detections nobody trusts; when a vendor questionnaire, SOC 2, ISO 27001, PCI, HIPAA, GDPR or NIS2 evidence is due; when a notification clock may already be running; when scoping an authorized test or a disclosure; or when a finding has to be written so an engineer, an executive or a board decides. Covers evidence handling, token eviction, and building a security program from nothing. Not for line-by-line secure code fixes (`security-best-practices`), STRIDE notation depth (`threat-modeling`), or implementing login flows (`auth`).
As a process A 83/100 · Runs to the end — weak spots: inputs and preconditions, execution cost, running it twice
What is at stake
Medium-severity findings: the skill is probably honest, but read what alarmed the scanner.
Below is the worst case for this category. The finding here is medium: the guard saw a sign, not a proof.
The skill asks for more than the task needs: broad tool access, credential environment variables, binaries. Every extra permission widens the damage from a mistake or a compromise.
Narrow allowed-tools and the variable list to the minimum; replace binaries with readable sources or scripts.
How to improve
- The SKILL.md body is over 5,000 tokens: move reference detail into references/ and load it when needed.
- Your own cases (evals/evals.json, 4–6 real requests with expected answers): the full check would then run those instead of a model-drafted suite.
- A spec.yaml with trigger phrases and assertions — a behaviour contract for CI; `skilltest init` writes a template.
Guard findings · 36
✓ No critical or high findings
Medium and low: 36
-
medium Broad scope
meta-agent-memory-dumpidentity.mdAgent memory / workspace files bundled with the skill (1) — likely a workspace dump with personal data or tokensidentity.md
-
low Risky intent
intent-offensive-securityauthorization.md:7Offensive-security / dual-use content (legitimate for authorised testing; review intended use) (quoted — discussed, not commanded)**Contents:** [The Three Questions](#the-three-questions) · [What Counts As Authorization](#what-counts-as-authorization) · [Rules Of Engagement](#rules-of-engagement) · [Fallbacks When Scope Is Missi
quoted -
low Risky intent
intent-offensive-securityauthorization.md:38Offensive-security / dual-use content (legitimate for authorised testing; review intended use)- **Techniques permitted**: exploitation or proof-of-existence only, credential attacks, lateral movement, social engineering, physical, denial of service. Each named, not implied.
-
low Risky intent
intent-offensive-securityauthorization.md:107Offensive-security / dual-use content (legitimate for authorised testing; review intended use)## Running A Bug Bounty Or VDP
-
low Risky intent
intent-offensive-securitycompliance.md:102Offensive-security / dual-use content (legitimate for authorised testing; review intended use)| Annual penetration test | Continuous testing and detection engineering find more | Do both; the annual test is the evidence, the continuous work is the security |
-
low Risky intent
intent-offensive-securitycompliance.md:123Offensive-security / dual-use content (legitimate for authorised testing; review intended use)Write it (`memory-template.md`): the scope boundary, in-scope systems, data classifications and retention minimums in `## Environment`; every audit finding, control gap and remediation as a `## Findin
-
low Risky intent
intent-offensive-securitydetection.md:23Offensive-security / dual-use content (legitimate for authorised testing; review intended use)| 2 | EDR telemetry | Execution, persistence and lateral movement on the endpoint |
-
low Risky intent
intent-offensive-securitydetection.md:96Offensive-security / dual-use content (legitimate for authorised testing; review intended use)| Credential access on the endpoint: LSASS handle, credential dumping tooling behaviour | EDR | Directly on the objective, not the tool |
-
low Risky intent
intent-offensive-securityforensics.md:28Offensive-security / dual-use content (legitimate for authorised testing; review intended use)- **Memory before disk, always**, on any host still running that matters. Memory holds the decrypted payload, the C2 configuration, the process ancestry, and — during a ransomware run — sometimes the
-
low Risky intent
intent-offensive-securityforensics.md:52Offensive-security / dual-use content (legitimate for authorised testing; review intended use)| Lateral movement out | RDP client artifacts, 3 network logons on the destination, `\\host\c$` access, remote service creation on the target |
-
low Risky intent
intent-offensive-securityincident-response.md:47Offensive-security / dual-use content (legitimate for authorised testing; review intended use) (quoted — discussed, not commanded)**Dwell time is why the window is wide.** Mandiant's M-Trends reports have tracked global median dwell time down into the sub-two-week range, but that median is dominated by ransomware, which announce
quoted -
low Risky intent
intent-offensive-securityincident-response.md:65Offensive-security / dual-use content (legitimate for authorised testing; review intended use)Never as containment: power off (destroys memory and often the encryption keys still resident during a ransomware run), reimage before capture, restore from backup before root cause, or "run the AV sc
-
low Risky intent
intent-offensive-securitymemory-template.md:18Offensive-security / dual-use content (legitimate for authorised testing; review intended use)| Recurring cadences: access review, key rotation, restore drill, tabletop, pentest, vendor reassessment, risk-acceptance expiry | `## Due` in `memory.md` | Fixed rows, dates updated |
-
low Risky intent
intent-offensive-securitymemory-template.md:128Offensive-security / dual-use content (legitimate for authorised testing; review intended use)- Ransomware playbook → `artifacts/playbook-ransomware.md`; read the moment encryption or extortion is suspected
-
low Risky intent
intent-offensive-securitymemory-template.md:166Offensive-security / dual-use content (legitimate for authorised testing; review intended use)| SEC-…019 | Build server reachable from the office VLAN on 22 | high | Lateral movement to the signing key | user | 2026-08-12 | open |
-
low Risky intent
intent-offensive-securitynetwork-security.md:40Offensive-security / dual-use content (legitimate for authorised testing; review intended use)| RDP (3389), SSH (22) with password authentication, VNC | Brute-forced and sprayed continuously; a favoured ransomware entry point |
-
low Risky intent
intent-offensive-securitynetwork-security.md:65Offensive-security / dual-use content (legitimate for authorised testing; review intended use) (detector / deny-list definition)- **The highest-value segment boundary in most organizations is user-workstation to user-workstation.** Workstations rarely need to talk to each other; blocking SMB and RDP between them removes the la
detector -
low Risky intent
intent-offensive-securitynetwork-security.md:68Offensive-security / dual-use content (legitimate for authorised testing; review intended use)- Flat networks with a firewall only at the perimeter are the norm and are the reason lateral movement is trivial. Any internal boundary is a substantial improvement over none.
-
low Risky intent
intent-offensive-securitynetwork-security.md:75Offensive-security / dual-use content (legitimate for authorised testing; review intended use)**Allowing all outbound 443 allows command and control by definition.** Nearly every intrusion needs outbound connectivity; almost nobody constrains it.
-
low Risky intent
intent-offensive-securityprogram.md:129Offensive-security / dual-use content (legitimate for authorised testing; review intended use)| The pentest as the programme | It samples one path at one moment; findings feed the backlog, they are not the plan |
-
low Risky intent
intent-offensive-securityprogram.md:137Offensive-security / dual-use content (legitimate for authorised testing; review intended use)Write it (`memory-template.md`): the current-state picture — crown jewels, systems, trust boundaries, log sources with their retention and gaps — in `## Environment`; every gap as a `## Findings` row
-
low Risky intent
intent-offensive-securityransomware.md:1Offensive-security / dual-use content (legitimate for authorised testing; review intended use)# Ransomware And Extortion
-
low Risky intent
intent-offensive-securityransomware.md:5Offensive-security / dual-use content (legitimate for authorised testing; review intended use) (quoted — discussed, not commanded)**Before anything**, read `incidents/<year>.md` and any `artifacts/playbook-ransomware.md` the `## Boxes` index in `~/Clawic/data/cybersecurity/memory.md` names — a playbook written calmly beats judge
quoted -
low Risky intent
intent-offensive-securityransomware.md:21Offensive-security / dual-use content (legitimate for authorised testing; review intended use)9. **Identify the family** from the note, the extension and a sample: ID Ransomware and the No More Ransom project both map notes and extensions to families, and free decryptors exist for a meaningful
-
low Risky intent
intent-offensive-securityransomware.md:116Offensive-security / dual-use content (legitimate for authorised testing; review intended use)After the incident, write it (`memory-template.md`): the incident row with awareness timestamp, encrypted scope, stolen-data assessment and outcome in `incidents/<year>.md`; the restore rate and the a
-
low Risky intent
intent-offensive-securitySKILL.md:5Offensive-security / dual-use content (legitimate for authorised testing; review intended use) (detector / deny-list definition)description: 'Runs defensive security: alert triage, compromise investigation, attack paths, vulnerability prioritization, detection, and risk reporting. Use when an alert, a suspicious login, a repor
detector -
low Risky intent
intent-offensive-securitySKILL.md:55Offensive-security / dual-use content (legitimate for authorised testing; review intended use) (detector / deny-list definition)- Deciding what to fix first: scanner output, a pentest report, a bug-bounty submission, an inherited backlog nobody has triaged
detector -
low Risky intent
intent-offensive-securitySKILL.md:67Offensive-security / dual-use content (legitimate for authorised testing; review intended use) (documentation table row)| Files encrypted, ransom note, extortion mail with stolen data | Do not reboot; size the encryption and the theft separately; validate a restore before any negotiation | `ransomware.md` |
table -
low Risky intent
intent-offensive-securitySKILL.md:71Offensive-security / dual-use content (legitimate for authorised testing; review intended use) (detector / deny-list definition)| Scanner output, a pentest report, thousands of CVEs and no plan | KEV / EPSS / exposure gate, then the SLA clock (Rules 4 and 6) | `vulnerabilities.md` |
detector -
low Risky intent
intent-offensive-securitySKILL.md:81Offensive-security / dual-use content (legitimate for authorised testing; review intended use) (detector / deny-list definition)| "Can I test this?", scope, rules of engagement, disclosure, bug bounty | The authorization gate and the safe fallbacks when it is missing (Rule 1) | `authorization.md` |
detector -
low Risky intent
intent-offensive-securityvulnerabilities.md:7Offensive-security / dual-use content (legitimate for authorised testing; review intended use)**Contents:** [The Gate, In Order](#the-gate-in-order) · [What Each Score Actually Means](#what-each-score-actually-means) · [Exposure Is The Multiplier](#exposure-is-the-multiplier) · [Triaging An In
-
low Risky intent
intent-offensive-securityvulnerabilities.md:87Offensive-security / dual-use content (legitimate for authorised testing; review intended use)## Reading A Pentest Report
-
low Risky intent
intent-offensive-securityvulnerabilities.md:89Offensive-security / dual-use content (legitimate for authorised testing; review intended use)- **A pentest samples one path at one moment.** Absence of findings is evidence about the test's scope and duration, not about the system. Read the scope and the constraints section first — that is wh
-
low Risky intent
intent-offensive-securityvulnerabilities.md:93Offensive-security / dual-use content (legitimate for authorised testing; review intended use)- Findings feed the backlog through the same gate as everything else. A pentest report is not a work plan and must never become the security programme (`program.md`).
-
low Risky intent
intent-offensive-securityvulnerabilities.md:118Offensive-security / dual-use content (legitimate for authorised testing; review intended use) (detector / deny-list definition)Write what the triage produced (`memory-template.md`): every finding that will be tracked as a row in `## Findings` with its id, severity, the attack path it removes, owner and due date — never a dupl
detector
A further 1 matches are quotations in this security skill's documentation and are not counted as findings.
Files scanned: 21. Evidence is masked. Grey chips explain why severity was lowered.
Against the Agent Skills spec
- warning
name-formatname should be kebab-case (lowercase letters, digits, hyphens) - warning
body-longSKILL.md body ≈ 8464 tokens (recommended < 5000); move details to references/ - note
frontmatter-keyunknown frontmatter key "slug" - note
frontmatter-keyunknown frontmatter key "homepage" - note
frontmatter-keyunknown frontmatter key "changelog"
Process rating: all ten parameters 83/100
- 0Inputs and preconditions. Does not say what the process needs to start
- 30Running it twice. 26 mutating operations with no state check
- 40Execution cost. Instruction body is 8464 tokens: crowds the task out of the window
- 100Tools and files. No external tools needed
- 100Steps. 51 steps
- 100Result and completion. Output format and completion criterion are stated
- 100When it triggers. States when to use and when not to
- 100Failures and branches. 7 branches, has a failure section
- 100Consistency. Name and required fields are in place
- 100Progress reporting. Reports progress
- medium Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- low 14 top-level sections: this looks like several domains in one skill
- low The response is described with custom markup (5 tags): a typed call is more reliable
Everything here is measured from the skill text rather than judged by a model, so the numbers are checkable. A parameter weighs more when it is a more common reason for the process to stall.
Quality signals
- +5Description has no quoted example phrases that should trigger the skill
- +3Description length 985: 120–800 characters recommended
- +4No input/output examples
- +1No license
- +2Single-language instructions
- +4Description says when NOT to use the skill
- +4Structure: 14 headings
- +3Step-by-step instructions: 51 items
- +3Output format is stated explicitly
Quality base 70; lint remarks subtract, signals add up to 100. Result: 66.