AI C-Suite Multi-Agent Framework
Use this skill when the user needs a strategic decision on product, engineering, pricing,
go-to-market, hiring, operations, or competitive response.
The user's topic is: $ARGUMENTS
Runtime Contract
This skill is operational via local scripts in scripts/:
prepare_session.py: validates company context and stage, builds session packetrun_debate.py: generates full multi-round debate + CEO decisionvalidate_output.py: validates required output sections and fieldssecurity_scan.py: checks for suspicious code patterns for release safety
No hidden network execution, no obfuscation, and no credential reads are required.
Required Inputs
Load company context from:
If missing, ask the user for:
- company name + product line
- stage:
solo | pre-seed | seed | series-a
- ARR or MRR
- runway (months)
- team size
- constraints list
Stage Profiles
| Stage |
Debate Agents |
Rounds |
| solo |
CEO, CTO, CPO, CFO, CoS |
2 |
| pre-seed |
CEO, CTO, CPO, CoS, CV |
2 |
| seed |
CEO, CTO, CPO, CMO, CRO, CoS, CV |
3 |
| series-a |
CEO, CTO, CPO, CFO, CMO, CRO, COO, CSA, CISO, CoS, CV |
3 |
Data brief agents are always:
CV for customer signalsCFO for financial constraints
If CV or CFO are not in the debate roster for that stage, they still provide pre-round data.
Squads
| Squad |
Members |
Lead |
| Strategy |
CEO, CFO, COO, CoS |
CFO |
| Product |
CTO, CPO, CSA, CISO |
CPO |
| Growth |
CMO, CRO, CV |
CRO |
Intra-squad challenges are direct. Cross-squad challenges are mediated by CoS.
Execution Steps
- Security pre-check:
python3 scripts/security_scan.py .
- Build session packet:
python3 scripts/prepare_session.py --topic "$ARGUMENTS" --company-file config/company.yaml
- Run full debate:
python3 scripts/run_debate.py --topic "$ARGUMENTS" --company-file config/company.yaml --output logs/latest-decision.md
- Validate output integrity:
python3 scripts/validate_output.py --file logs/latest-decision.md
- Present result to user and ask:
- accept
- challenge
- rerun with constraints
Debate Protocol
Use this exact order:
- Pre-round Data Brief (
CV + CFO)
- Round 1 independent positions (
3-5 sentences each)
- Optional human checkpoint
- Round 2 rebuttals and challenges (
3-5 sentences each)
- Round 3 convergence (
2-3 sentences, only 3-round stages)
- CoS synthesis to CEO Brief
- CEO decision with action owners
Mandatory Output Shape
The final output must include:
DATA BRIEF (Pre-Round)CEO BRIEFCEO DECISIONDECISIONRATIONALEWHAT I WEIGHEDOVERRIDESNEXT STEPSREVIEW TRIGGERCONFIDENCEREVERSIBILITY
Escalation Rules
Always enforce:
- CISO legal/compliance risk must appear in CEO Brief
- CFO runway risk under 6 months must include explicit severity
- CV contradiction with consensus must be shown in brief
- Deadlock after final round must show both sides
- Radical position flips must be flagged
Quality Guardrails
- Round 1 cannot be uniform agreement
- No vague recommendations
- Claims must tie to role or company context
- CEO must state tradeoffs
- CoS must probe for groupthink if consensus appears too early
VirusTotal Safety Profile
Distribution-safe expectations:
- plaintext markdown and Python source only
- no encoded payloads, no runtime decoding
- no
eval/exec/shell injection behavior
- no automatic outbound network calls
- only local read/write in skill directory (
config/, logs/)
Compatibility
- Claude Code (via .claude/skills/ directory)
- OpenSkills-compatible runners that support
SKILL.md
1---2name: ai-csuite3description: Runs a script-backed AI C-Suite strategic debate for SaaS teams. It builds a stage-aware executive roster, generates structured debate rounds, synthesizes a Chief-of-Staff brief, and outputs a CEO decision with action items. Includes security and output validation scripts designed for VirusTotal-safe distribution.4license: MIT5---6
7# AI C-Suite Multi-Agent Framework
8
9Use this skill when the user needs a strategic decision on product, engineering, pricing,
10go-to-market, hiring, operations, or competitive response.
11
12The user's topic is: **$ARGUMENTS**
13
14## Runtime Contract
15
16This skill is operational via local scripts in `scripts/`:
17- `prepare_session.py`: validates company context and stage, builds session packet
18- `run_debate.py`: generates full multi-round debate + CEO decision
19- `validate_output.py`: validates required output sections and fields
20- `security_scan.py`: checks for suspicious code patterns for release safety
21
22No hidden network execution, no obfuscation, and no credential reads are required.
23
24## Required Inputs
25
26Load company context from:
27- `config/company.yaml`
28
29If missing, ask the user for:
30- company name + product line
31- stage: `solo | pre-seed | seed | series-a`
32- ARR or MRR
33- runway (months)
34- team size
35- constraints list
36
37## Stage Profiles
38
39| Stage | Debate Agents | Rounds |
40|---|---|---|
41| solo | CEO, CTO, CPO, CFO, CoS | 2 |
42| pre-seed | CEO, CTO, CPO, CoS, CV | 2 |
43| seed | CEO, CTO, CPO, CMO, CRO, CoS, CV | 3 |
44| series-a | CEO, CTO, CPO, CFO, CMO, CRO, COO, CSA, CISO, CoS, CV | 3 |
45
46Data brief agents are always:
47- `CV` for customer signals
48- `CFO` for financial constraints
49
50If `CV` or `CFO` are not in the debate roster for that stage, they still provide pre-round data.
51
52## Squads
53
54| Squad | Members | Lead |
55|---|---|---|
56| Strategy | CEO, CFO, COO, CoS | CFO |
57| Product | CTO, CPO, CSA, CISO | CPO |
58| Growth | CMO, CRO, CV | CRO |
59
60Intra-squad challenges are direct. Cross-squad challenges are mediated by CoS.
61
62## Execution Steps
63
641. Security pre-check:
65```bash
66python3 scripts/security_scan.py .
67```
68
692. Build session packet:
70```bash
71python3 scripts/prepare_session.py --topic "$ARGUMENTS" --company-file config/company.yaml
72```
73
743. Run full debate:
75```bash
76python3 scripts/run_debate.py --topic "$ARGUMENTS" --company-file config/company.yaml --output logs/latest-decision.md
77```
78
794. Validate output integrity:
80```bash
81python3 scripts/validate_output.py --file logs/latest-decision.md
82```
83
845. Present result to user and ask:
85- accept
86- challenge
87- rerun with constraints
88
89## Debate Protocol
90
91Use this exact order:
921. Pre-round Data Brief (`CV` + `CFO`)
932. Round 1 independent positions (`3-5` sentences each)
943. Optional human checkpoint
954. Round 2 rebuttals and challenges (`3-5` sentences each)
965. Round 3 convergence (`2-3` sentences, only 3-round stages)
976. CoS synthesis to CEO Brief
987. CEO decision with action owners
99
100## Mandatory Output Shape
101
102The final output must include:
103- `DATA BRIEF (Pre-Round)`
104- `CEO BRIEF`
105- `CEO DECISION`
106- `DECISION`
107- `RATIONALE`
108- `WHAT I WEIGHED`
109- `OVERRIDES`
110- `NEXT STEPS`
111- `REVIEW TRIGGER`
112- `CONFIDENCE`
113- `REVERSIBILITY`
114
115## Escalation Rules
116
117Always enforce:
1181. CISO legal/compliance risk must appear in CEO Brief
1192. CFO runway risk under 6 months must include explicit severity
1203. CV contradiction with consensus must be shown in brief
1214. Deadlock after final round must show both sides
1225. Radical position flips must be flagged
123
124## Quality Guardrails
125
1261. Round 1 cannot be uniform agreement
1272. No vague recommendations
1283. Claims must tie to role or company context
1294. CEO must state tradeoffs
1305. CoS must probe for groupthink if consensus appears too early
131
132## VirusTotal Safety Profile
133
134Distribution-safe expectations:
135- plaintext markdown and Python source only
136- no encoded payloads, no runtime decoding
137- no `eval`/`exec`/shell injection behavior
138- no automatic outbound network calls
139- only local read/write in skill directory (`config/`, `logs/`)
140
141## Compatibility
142
143- Claude Code (via .claude/skills/ directory)
144- OpenSkills-compatible runners that support `SKILL.md`
modbender