AB mid-engagement-ir-detection
Methodology for detecting client SOC patches, attacker activity, and security-state changes that occur DURING a red-team engagement — and converting those observations into deliverable findings. Built from authorized red-team work where the client patched a confirmed SQLi within 30 minutes of detection AND an external attacker locked multiple new accounts during a single test session. Use when (a) running ANY active engagement against a monitored target, (b) a previously-confirmed finding stops reproducing, (c) baseline timing shifts unexpectedly, or (d) you notice response patterns changing during testing.
Methodology for detecting client SOC patches, attacker activity, and security-state changes that occur DURING a red-team engagement — and converting those…
As a process B 70/100 · Nearly there — weak spots: result and completion, inputs and preconditions, running it twice
How to improve
- 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 · 3
✓ No critical or high findings
Medium and low: 3
-
low Risky intent
intent-offensive-securitySKILL.md:20Offensive-security / dual-use content (legitimate for authorised testing; review intended use)- Bug bounty (client doesn't know you're there; no real-time IR)
-
low Risky intent
intent-offensive-securitySKILL.md:304Offensive-security / dual-use content (legitimate for authorised testing; review intended use)This is the discipline that distinguishes professional red team from "hobbyist scanning". Your client wants the timeline of vulnerability + mitigation, not just the static state.
-
low Risky intent
intent-offensive-securitySKILL.md:337Offensive-security / dual-use content (legitimate for authorised testing; review intended use)- `bb-methodology` — note that this skill is INAPPROPRIATE for bug bounty (no real-time IR there)
Files scanned: 1. Evidence is masked. Grey chips explain why severity was lowered.
Against the Agent Skills spec
- note
frontmatter-keyunknown frontmatter key "sources" - note
frontmatter-keyunknown frontmatter key "report_count"
Process rating: all ten parameters 70/100
- 0Result and completion. Does not say what the result is
- 0Inputs and preconditions. Does not say what the process needs to start
- 30Running it twice. 10 mutating operations with no state check
- 70Execution cost. Instruction body is 4163 tokens
- 100Tools and files. No external tools needed
- 100Steps. 66 steps
- 100When it triggers. States when to use and when not to
- 100Failures and branches. 1 branches, has a failure section
- 100Consistency. Name and required fields are in place
- 100Progress reporting. Reports progress
- low 12 top-level sections: this looks like several domains in one skill
- high The skill tells the model to perform an irreversible action with no human approval
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
- +3Output format is not stated: the model decides each time
- +1No license
- +2Single-language instructions
- +4Description says when NOT to use the skill
- +3Description length 614: enough signal without eating the budget
- +4Structure: 21 headings
- +3Step-by-step instructions: 66 items
- +4Has examples (10 code blocks)
Quality base 70; lint remarks subtract, signals add up to 100. Result: 86.