company-skill 4.0.0 → 4.2.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (2) hide show
  1. package/README.md +118 -85
  2. package/package.json +1 -1
package/README.md CHANGED
@@ -1,91 +1,137 @@
1
- # Company
1
+ # /company
2
2
 
3
- Give it a goal. The whole company works until it's done.
3
+ [![npm](https://img.shields.io/npm/v/company-skill)](https://www.npmjs.com/package/company-skill) [![license](https://img.shields.io/npm/l/company-skill)](LICENSE) [![downloads](https://img.shields.io/npm/dw/company-skill)](https://www.npmjs.com/package/company-skill)
4
+
5
+ > *You don't prompt agents one at a time. You write a team in markdown, hand them a goal, and go to sleep. In the morning, STATUS.md tells you what got done, what got rejected, and what the company learned. The playbook from session 3 makes session 4 faster. By session 10, the company runs itself better than you could direct it manually.*
6
+
7
+ **Define your team in markdown. Give it a goal. Walk away.**
8
+
9
+ A Claude Code skill that runs your entire company — CEO delegates, departments execute in parallel, built-in reviewers verify — and doesn't stop until the goal is done.
4
10
 
5
11
  ```
6
12
  /company "Build the user auth system with OAuth2"
7
13
  ```
8
14
 
9
- A Claude Code skill that reads your team structure from `COMPANY.md`, runs every employee in loops, and doesn't stop until built-in reviewers verify the goal is met.
15
+ ## Why /company
10
16
 
11
- ## Install
17
+ | | Without /company | With /company |
18
+ |---|---|---|
19
+ | Task routing | You manually prompt each agent | CEO reads the goal, picks relevant employees, delegates |
20
+ | Quality gates | Hope it's correct | Reviewer + Devil's Advocate + Elegance Enforcer triple-check |
21
+ | Knowledge retention | Lost every session | Playbook accumulates what worked, what failed, what's faster |
22
+ | Parallelism | One agent at a time | All departments run in parallel |
23
+ | Stopping condition | You decide when it's done | criteria.json blocks exit until ALL criteria pass |
12
24
 
25
+ ## Quick Start
26
+
27
+ **1. Install**
13
28
  ```bash
14
29
  npx company-skill install
15
30
  ```
16
31
 
17
- Or from git:
18
- ```bash
19
- curl -sL https://raw.githubusercontent.com/jagmarques/company-skill/main/install.sh | bash
32
+ **2. Define your team** (optional — a minimal company is created automatically)
33
+ ```markdown
34
+ ## Engineering
35
+ - Backend Lead, API design and database architecture
36
+ - Frontend Dev, React components and state management
37
+
38
+ ## Research
39
+ - ML Scientist, model experiments and benchmarks
20
40
  ```
21
41
 
22
- Edit `COMPANY.md` with your team. Or skip it, the skill creates a minimal company automatically.
42
+ **3. Run**
43
+ ```
44
+ /company "Build a REST API for user management with tests"
45
+ ```
23
46
 
24
47
  ## How It Works
25
48
 
49
+ ```mermaid
50
+ graph LR
51
+ G[GOAL] --> T[THINK]
52
+ T -->|Opus: CEO + leads assign tasks| E[EXECUTE]
53
+ E -->|Sonnet: workers do the work| V[VERIFY]
54
+ V -->|Opus: Reviewer + Advocate| D{Done?}
55
+ D -->|NO: feedback| T
56
+ D -->|YES| S[STATUS.md]
26
57
  ```
27
- GOAL: "Build the auth system"
28
58
 
29
- THINK CEO picks relevant employees, leads assign tasks
30
- EXECUTE Employees do the work, use installed skills
31
- VERIFY Reviewer checks criteria.json, Advocate attacks results
59
+ The loop does NOT stop until the Reviewer confirms all criteria pass AND the Devil's Advocate accepts. There is no iteration limit.
32
60
 
33
- Not done? Loop back with feedback.
34
- Done? Update playbook, write STATUS.md.
35
- ```
61
+ <details>
62
+ <summary><strong>THINK</strong> CEO picks relevant employees, leads assign tasks</summary>
63
+
64
+ The CEO reads the goal and COMPANY.md, decides which departments and employees are relevant (a mobile app goal doesn't need a Topologist), writes an active roster, then launches all department leads in parallel. Each lead assigns tasks to their employees with one sentence, one skill, and context.
65
+
66
+ If a lead sees a skill gap, they write `HIRE: {role}, {why}` and the CEO adds it to the team.
67
+ </details>
68
+
69
+ <details>
70
+ <summary><strong>EXECUTE</strong> — All workers run in parallel with installed skills</summary>
36
71
 
37
- The loop runs until ALL criteria in `criteria.json` pass. A Stop Hook blocks Claude from exiting early. To cancel: `touch .company/CANCEL`.
72
+ Every employee gets their task, previous findings, and failed approaches from the playbook. Every finding must have a source — file path, URL, or command output. Novel ideas use "NOVEL needs validation" and the reviewer adds a validation criterion. No source = rejected.
73
+ </details>
74
+
75
+ <details>
76
+ <summary><strong>VERIFY</strong> — Triple quality gate blocks premature completion</summary>
77
+
78
+ **Internal Reviewer** checks each criterion in criteria.json against evidence. No evidence? Stays `false`.
79
+
80
+ **Devil's Advocate** attacks anything marked as passing. "Is this actually complete or surface-level? What edge cases were missed?"
81
+
82
+ **Elegance Enforcer** asks "Can this be simpler? Does every component justify its existence?"
83
+
84
+ All three must accept before the loop exits.
85
+ </details>
38
86
 
39
87
  ## Goal Enforcement
40
88
 
41
- The skill creates `criteria.json` with checkable success criteria:
89
+ The skill creates `criteria.json` with machine-checkable success criteria:
42
90
 
43
91
  ```json
44
- {"goal":"Build auth","criteria":[
45
- {"id":1,"description":"OAuth2 login works with Google","passes":false,"evidence":null},
46
- {"id":2,"description":"All tests pass","passes":false,"evidence":null}
92
+ {"goal": "Build auth", "criteria": [
93
+ {"id": 1, "description": "OAuth2 login works with Google", "passes": false, "evidence": null},
94
+ {"id": 2, "description": "All tests pass", "passes": false, "evidence": null}
47
95
  ]}
48
96
  ```
49
97
 
50
- The reviewer updates `passes` to `true` with evidence as work completes. The stop hook reads this file and blocks exit until everything passes.
98
+ A Stop Hook reads this file and **blocks Claude from exiting** until every criterion passes. To cancel: `touch .company/CANCEL`.
51
99
 
52
- ## Self-Improvement
100
+ ## Self-Improving Playbook
53
101
 
54
102
  One file: `.company/playbook.md`. Accumulates across sessions.
55
103
 
56
- After each session, the CEO writes:
57
- - WORKED: what succeeded (with evidence)
58
- - FAILED: what failed, USE INSTEAD: what works, WHY: the difference
59
- - INEFFICIENT: what was slow, FASTER: better approach
60
- - TOP: best employees for priority activation next time
61
- - HIRE/FIRE: roles added or deactivated
104
+ After each session, the CEO writes what worked, what failed (and what to use instead), what was slow (and what's faster), which employees performed best, and which roles to hire or deactivate. Leads read the playbook before every THINK phase.
62
105
 
63
- Leads read the playbook before every THINK phase. Employees check failed approaches before proposing new ones. The company that starts session 5 is smarter than session 1.
106
+ **The company that starts session 5 is smarter than session 1.**
64
107
 
65
- The CEO also updates COMPANY.md: tags `[inactive]` on zero-contribution roles, `[priority]` on top performers, adds hired roles, evolves employee descriptions based on what they're good at.
108
+ The CEO also evolves COMPANY.md: tags `[inactive]` on zero-contribution roles, `[priority]` on top performers, and updates employee descriptions based on what they're actually good at.
66
109
 
67
110
  ## Built-In Roles
68
111
 
69
- Every company gets these automatically:
112
+ Every company gets these automatically (deduplicated if you define them in COMPANY.md):
70
113
 
71
- | Role | Phase | What they do |
72
- |------|-------|-------------|
73
- | CEO | THINK | Picks relevant employees for the goal, resolves conflicts |
74
- | CTO | THINK | Technical decisions, architecture |
114
+ | Role | Phase | Purpose |
115
+ |------|-------|---------|
116
+ | CEO | THINK | Reads goal, picks relevant employees, resolves conflicts |
117
+ | CTO | THINK | Technical decisions, architecture review |
75
118
  | Internal Reviewer | VERIFY | Checks criteria.json, rejects findings without sources |
76
- | User Advocate | VERIFY | Represents the end user |
77
- | Devil's Advocate | VERIFY | Attacks results, finds holes |
78
- | Elegance Enforcer | VERIFY | Prevents over-engineering |
119
+ | User Advocate | VERIFY | "Would a real user understand this?" |
120
+ | Devil's Advocate | VERIFY | Attacks results, finds holes, prevents false completion |
121
+ | Elegance Enforcer | VERIFY | Prevents over-engineering, kills unnecessary complexity |
79
122
 
80
- Deduplicated if you define them in COMPANY.md.
123
+ A 2-person COMPANY.md (Backend Dev + Frontend Dev) automatically gets CEO + CTO + both devs + all 4 reviewers = **8 employees running**.
81
124
 
82
- ## Source Citations
125
+ ## Model Assignment
83
126
 
84
- Every finding needs a source:
85
- - Existing claims: file path, URL, or command output
86
- - Novel ideas: "NOVEL - needs validation" (reviewer adds a validation criterion)
127
+ | Phase | Model | Who |
128
+ |-------|-------|-----|
129
+ | THINK | Opus | CEO, CTO, department leads |
130
+ | EXECUTE | Sonnet | Workers |
131
+ | VERIFY | Opus | All reviewers |
132
+ | COMPRESS | Haiku | Digest writer |
87
133
 
88
- No source = rejected by reviewer.
134
+ Override per employee: `- ML Scientist, experiments [opus]`
89
135
 
90
136
  ## Commands
91
137
 
@@ -97,66 +143,32 @@ No source = rejected by reviewer.
97
143
  /company:resume Continue from last session
98
144
  ```
99
145
 
100
- ## Visual Indicators
101
-
102
- ```
103
- ════════════════════════════════════════════════
104
- CYCLE 1 - THINK > EXECUTE > VERIFY
105
- ════════════════════════════════════════════════
106
-
107
- CYCLE 1 VERDICT: NOT DONE
108
- Missing validation of compression ratios
109
- ```
110
-
111
- Employees show with colors: leads (cyan), workers (green), reviewers (yellow), digest (gray).
112
-
113
- ## Agents
114
-
115
- | Agent | Phase | Color |
116
- |-------|-------|-------|
117
- | company-lead | THINK | Cyan |
118
- | company-worker | EXECUTE | Green |
119
- | company-reviewer | VERIFY | Yellow |
120
- | company-critic | VERIFY | Yellow |
121
- | company-digest | COMPRESS | Gray |
122
-
123
- ## Model Assignment
124
-
125
- | Phase | Model | Who |
126
- |-------|-------|-----|
127
- | THINK | Opus | CEO, CTO, leads |
128
- | EXECUTE | Sonnet | Workers |
129
- | VERIFY | Opus | Reviewers |
130
- | COMPRESS | Haiku | Digest writer |
131
-
132
- Override per employee: `- ML Scientist, experiments [opus]`
133
-
134
146
  ## Installed Skills
135
147
 
136
- Auto-installed on first run:
148
+ Auto-installed on first run. When installed, employees MUST use them.
137
149
 
138
150
  | Pack | What employees get |
139
151
  |------|-------------------|
140
- | gstack | /review, /ship, /qa, /investigate, /browse, /office-hours |
141
- | GSD | /gsd:plan-phase, /gsd:execute-phase, /gsd:verify-work, /gsd:debug |
152
+ | gstack | /review, /ship, /qa, /investigate, /browse |
153
+ | GSD | /gsd-plan-phase, /gsd-execute-phase, /gsd-verify-work, /gsd-debug |
142
154
  | trailofbits | Security audit, vulnerability detection |
143
155
 
144
- Install manually for more:
156
+ <details>
157
+ <summary>Install more skill packs</summary>
145
158
 
146
159
  ```
147
160
  /plugin marketplace add obra/superpowers-marketplace
148
161
  /plugin marketplace add wshobson/agents
149
162
  /plugin marketplace add alirezarezvani/claude-skills
150
163
  ```
151
-
152
- When installed, employees MUST use them.
164
+ </details>
153
165
 
154
166
  ## What Gets Created
155
167
 
156
168
  ```
157
169
  .company/
158
170
  criteria.json Machine-checkable goal state
159
- playbook.md Accumulated lessons (self-improvement)
171
+ playbook.md Accumulated lessons (THE self-improvement file)
160
172
  active-roster.md Employees activated for this goal
161
173
  active-tasks.md Deduplicated task list
162
174
  STATUS.md Final report
@@ -165,6 +177,27 @@ When installed, employees MUST use them.
165
177
  {dept}/ Per-employee findings (persist across sessions)
166
178
  ```
167
179
 
180
+ ## Design Choices
181
+
182
+ Three principles behind the skill:
183
+
184
+ - **One file to define the team.** COMPANY.md is the only thing you write. Everything else — delegation, task routing, quality checks — is automatic.
185
+ - **No iteration limit.** The loop runs until criteria.json says done. Not 3 cycles. Not 5. Until the Reviewer and Devil's Advocate both accept.
186
+ - **Self-improvement over configuration.** Instead of tuning prompts, the company learns from its own failures. The playbook accumulates across sessions. Roles get tagged `[priority]` or `[inactive]` based on performance. The system gets better by running, not by tweaking.
187
+
188
+ ## Project Structure
189
+
190
+ ```
191
+ COMPANY.md Your team definition (the only file you edit)
192
+ skill/SKILL.md The skill logic (THINK > EXECUTE > VERIFY loop)
193
+ agents/ Subagent definitions (lead, worker, reviewer, critic, digest)
194
+ hooks/ Stop guard, session restore, precompact
195
+ commands/ run.md, resume.md, status.md
196
+ examples/ Sample team configurations
197
+ install.sh Curl-based installer
198
+ bin/install.js npx installer
199
+ ```
200
+
168
201
  ## Examples
169
202
 
170
203
  | File | Team |
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "company-skill",
3
- "version": "4.0.0",
3
+ "version": "4.2.0",
4
4
  "description": "Goal-driven multi-employee company for Claude Code. Give it a goal, it runs until done.",
5
5
  "bin": {
6
6
  "company-skill": "./bin/install.js"