Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add commands/arcassystems/founderos/devilgit clone --depth 1 https://github.com/ARCASSystems/FounderOSWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00092 | $0.00636 |
| Opus 5 | $0.00046 | $0.00318 |
| Sonnet 5 | $0.00018 | $0.00127 |
| Haiku 4.5 | $0.00009 | $0.00064 |
Grade A, and why
devil scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 49 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Devil's advocate
Turn the bias lens on the position itself, not on the person holding it. The target is the strongest honest case against, plus an honest confidence read. This runs the output self-check in rules/biases.md on demand.
Argument: $ARGUMENTS (the claim, recommendation, or decision to challenge). If empty, challenge the most recent opinion in this conversation. If there is none, reply What should I challenge? Re-run as /founder-os:devil <claim>. and stop.
Procedure
- Read
rules/biases.mdfor the six output biases and the contract. - If the claim touches live state (a lead, a decision, a relationship, a spend), read the relevant context file first so the challenge is grounded, not generic. Read
brain/.snapshot.mdif it exists. - Name which one of the six biases is most likely loading the original position. One, not all six.
- Build the counter. Do not soften it. Do not perform balance. The job here is the other side.
Output format
DEVIL: <one-line restatement of the position being challenged>
MOST LIKELY BIAS: <one of the six> - <one line on how it shows up here>
THE COUNTER-CASE
-> <strongest argument against, specific>
-> <second, only if there is a real one>
CONFIDENCE: high / medium / low - <the one fact that would move it>
ABSENT: <what evidence is missing that would change the call>
IF YOU DO NOTHING: <what actually happens if this is not acted on>
HONEST READ: <does the position survive the challenge or not, one line, no hedging>
Rules
- Challenge the position, never the person. No "you're being biased".
- If the position survives the challenge, say so plainly. The goal is truth, not a kill.
- Anti-theater: if you cannot build a real counter, say "low confidence, thin evidence", never "all clear".
- No em dashes, no en dashes. No banned words. No rule-of-three.
Cross-refs
- Runs:
rules/biases.md(the output bias self-check) - Companion to
/founder-os:forcing-questions(whether to START something) and thedecision-frameworkskill (choosing between options). This one stress-tests a position already on the table.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 49 lines · 92 tokens per session scan A d35d426f614c
devil is a command published in the GitHub repository ARCASSystems/FounderOS (5 stars, last pushed 18d ago), licensed MIT. It adds 92 tokens to every session and 636 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other commands, from other repositories
wiki-query
Ask questions against the wiki. Synthesizes answers from wiki pages with cross-reference citations.
wiki-digest
Generate a daily or weekly digest of wiki changes — new sources, pages, insights, and gaps.
demo-command
Example slash command that wraps the demo-skill. Showcases the command kind.
ingest-l1
L1 analysis loop for the abapwiki knowledge base: for each batch it launches the abap-analyzer sub-agent in parallel, then the adversarial judge abap-deepcheck (separate session), applies only the analyses that pass the fail-closed gate, and commits. Resumes exactly after an interruption. Use this skill to document…
web-search
Search fetched web-source Markdown by regex and pull surrounding context for the best hits. Optionally restrict to one source alias.
docs-refresh
Re-fetch a cached source, ignoring the 7-day cache. Use when upstream content has changed.