Use when about to take any outward or offensive action (request, payload, persistence, lateral movement, exfil, or feeding captured traffic to the model) — to decide detection footprint, cleanup, and secret redaction first
SkillScore83Maintained
Automatic
◆Security & AppSec pack
Last verified 8/21/2026
What it changes
Passed 19 of 23 test cases with the skill, and 14 of 23 without it.
Without skill61%
With skill83%
Checks it passed that the model alone missed
Names redact_text() as the sanitization function
Mentions using redact_headers.py or redact_text()
Selects the lower-noise execution variant if available
and 3 more in the full report
5 fixed14 passed either way4 still failing0 broke
Pass-rate lift
+22 pp
Tokens
-4%
Agent turns
0%
Response time
-29%
Model gemini-3.6-flashMeasured 8/21/2026v119 of 23 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 5 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 5
Prompt
In Python code processing API responses before sending them to model context, what function call from the coding mastery library performs inline text redaction of sensitive headers?
What was checked, without and with the skill
Checked
Without skill
With skill
Names redact_text() as the sanitization function
Missed
Passed
case-22 · gemini-3.6-flash · 8/21/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md37 linessingle file
OPSEC Discipline
Overview
Decide the detection footprint and cleanup before acting, not after. Every offensive technique
in this repo pairs with the telemetry it generates and a detection signature — use that pairing to
choose the quieter path and to know what you'll need to clean up.
Before acting, answer three questions
Detection — what telemetry does this generate? Which Sigma/EDR rule would catch it? (Each
domain skill's OPSEC & Detection table has this.) Pick the lower-noise variant when one exists.
Cleanup — what does this touch (files, registry, services, accounts, tickets)? How is it
reverted? Stage the cleanup before you create the artifact.
Secret hygiene — any captured traffic / output that reaches the model or the report must be
redacted at the boundary:
python skills/coding-mastery/scripts/_lib/redact_headers.py < exchange.txt (or redact_text()),
so Authorization/Cookie/API-key/JWT values never land in context or the report.
Red Flags — STOP, decide OPSEC first
"I'll worry about cleanup after I'm in" (you won't; decide now)
"Just paste the raw request/response" (redact secrets at the boundary first)
"Loudest exploit is fine, it works" (pick the variant with the smaller footprint)
"Persistence now, document later" (persistence on un-authorized hosts is out — see scope-discipline)
"The token's only in a log" (a logged secret is a leaked secret — mask it)
Rationalizations
Excuse
Reality
"OPSEC slows me down"
Unplanned noise/loot is how engagements get burned and how data leaks.
"It's an internal tool log"
Logs get shipped, shared, and indexed. Redact.
"Cleanup is a reporting-phase task"
You can't clean what you didn't track. Track at action time.
OPSEC is part of every action, not a phase. Detection-awareness also strengthens the defensive value
of the finding (you can tell the client exactly how to catch it).