Enforce Red-Team verification and adversarial protocol audit. Use only when verifying completion, performing self-scans, or checking protocol violations; do not activate for ordinary implementation, configuration, or unit-test requests.
699 tokens
context cost
the whole folder, loaded on every use
1
files
instructions only
0
copies elsewhere
how many repositories repackaged it
536
stars on the repo
on the repository, not the skill itself
Install
one command, takes just this skill from the repository
Before declaring any task "done" or calling notify_user:
Adversarial Audit: Search for Standard Defaults where project rules should exist.
Protocol Check: Confirm active skills and workflows were loaded before writing.
Evidence Check: Ask what command or artifact proves the completion claim.
Execution Bias Check: Ask whether speed or convenience skipped a structural rule.
Post-Write Self-Scan
Immediately after tool call:
Scan: Read diff or file content.
Match: Check against Anti-Patterns in all active skills.
Fix: Re-edit immediately if violation detected.
Red Flags
Stop if "done" appears before fresh verification: No completion claim yet.
Stop if you relied on memory instead of re-reading files: Reload source of truth.
Stop if the shortcut is "small enough to skip protocol": Small changes hide drift.
Rationalization Prevention
"The change is tiny": Tiny changes still violate guardrails.
"The test passed earlier": Old evidence does not prove current state.
"I know the pattern already": Load the active skill anyway.
Anti-Patterns
No "Done" Bias: Functional success != Protocol success.
No Reliance on Memory: Always retrieval-led (Skill view_file) before write.
No Skipping Protocols: "Small changes" where most violations happen.
Execution Bias Detection
Look for:
Local mocks instead of shared fakes.
Hardcoded styles instead of design tokens.
Try-catch blocks without standard error handling.
Missing Pre-Write Audit Log in thoughts.
References
Canonical response anchors
When this skill applies, preserve the following domain terminology or equivalent concrete examples in the answer when relevant:
active skill
active skills
reload
Additional task-grounded exact anchors: Pre-Write Audit; Standard Defaults; Anti-Patterns
When reporting a UI or implementation self-scan, explicitly check for hardcoded styles and missing design tokens, and name the Pre-Write Audit evidence before accepting the result.