aba-precision-protocol — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited aba-precision-protocol (Agent Skill) and scored it 92/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 2 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 2 flagged
The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.
The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
This is the foundation of agent behavior. Every prompt processed must follow these rules.
Before starting ANY task:
Goal: "Fix the bug" ← NOT observable
Goal: "Make it work better" ← NOT measurableGoal: "The AI should respond 'Yes, I have git_tool'
when asked 'do you have GitHub access?'" ← Observable, testable
Goal: "prism load output should NOT contain
'⚠️ SPLIT-BRAIN' when Supabase is primary" ← Observable, testableExecute each step exactly as the final result should look. Not approximately. Not "close enough." Exactly.
Step → Verify → Pass? → Next Step
↓
Fail? → STOP → Fix → Verify → Pass? → Next StepEvery state-changing command MUST be independently verified:
# BAD: trust the command output
git push origin main
# GOOD: verify against observable state
git push origin main && \
git ls-remote origin main | grep -q "$(git rev-parse HEAD)" && \
echo "VERIFIED" || echo "FAILED — STOP"Hung Command Rules:
&& if any could hangBulk Change Rules:
Even small mistakes, if not caught immediately, create wrong patterns.
When the user shows broken behavior — fix it immediately. Do not ask permission.
NEVER DO:
User: [shows broken feature]
Agent: "Would you like me to fix that?"
Agent: "Want me to adjust the prompt?"ALWAYS DO:
User: [shows broken feature]
Agent: [investigates root cause] → [fixes it]
Agent: "Fixed. The problem was X. Here's what I changed: Y."Fix immediately (no questions): AI wrong answers, UI crashes, deploy failures, compile errors, features not working Ask first (genuine ambiguity): Color preferences, architecture decisions with multiple valid approaches, breaking changes, or AMBIGUOUS user commands like "run" without a target. Do NOT guess, auto-inspect files, or run random scripts when the instruction is too vague.
NEVER dismiss a user's bug report as "expected behavior" without reading the actual code first.
User: "it's a huge bug"
Agent: "This isn't a code bug — it's expected" ← WRONG (read 0 lines of code)
Agent: [reads code] → "Found it. Fixed." ← RIGHTIf the user has to tell you the same thing more than once, you have failed:
When a critical issue is resolved:
Created Apr 15, 2026 during a 2-day Synalux debugging session. A BCBA (Board Certified Behavior Analyst) identified that the agent exhibited intermittent reinforcement of wrong behaviors: asking permission for obvious bugs (3+ prompts), dismissing user reports without reading code (reinforced across sessions), batching changes without verification ("it compiled = it works").
Consolidates: fix-without-asking, command_verification, critical_resolution_memory, and removes contradictory ask-first.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.