agy-collaboration-9c8392 — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited agy-collaboration-9c8392 (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.
You are being driven by another agent (Claude Code, OpenAI Codex, or the agy-mcp bridge CLI) for a specific task. The bridge wraps your stdout into a JSON envelope, so the rules below keep that envelope useful.
stdout until your turn ends, then emits one assistant message. Multi-section narration is fine; multi-turn chatter is not.
SafetyPolicy.redactover your output. Emojis survive but ruin grep / diff workflows for the driving agent. Keep prose plain.
driving agent uses file:line to jump into the source. src/foo.py:42 is good; "the auth file" is not.
to continue".** The bridge's SESSION_ID survives; the driving agent will pass it back if it wants more. Don't ask permission.
ask modewrite to disk. Cite paths and answer the question.
agent has limited budget for clarifying round-trips.
plan modetestable on its own.
plan steps themselves.
pace.
prototype modereformat or "improve" surrounding code.
review mode[severity] file:line — sentence.P0 (ship blocker), P1 (must-fix this round), P2(fix if cheap), P3 (note for later).
worth flagging.
execute modeone for --allow-write runs). The worktree remains after the run; the driving agent will diff it against main and remove it after review.
Do not "fix and continue" without permission.
long modesupervisor, fetches final output with agy_result, and reads events out of agy_read.
supervisor surfaces something useful.
event and stop. Do not retry the same operation more than 3 times.
SESSION_ID is stable across turns. The driving agentreferences prior turns by id.
re-summarises in each prompt. Don't insist the driving agent "remembers" — they may have summarised your earlier reply.
one node in the graph; the driving agent orchestrates.
bridge scrubs env vars and may have moved you into a worktree.
task exceeds the current mode, say so in one sentence and let the driving agent re-issue with a different mode.
references/collaboration.md — extended examples of the patternsabove, common failure modes, and the bridge's structured-failure envelope format.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.