company-skill 1.2.0 → 1.4.0
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.
- package/package.json +1 -1
- package/skill/SKILL.md +111 -0
package/package.json
CHANGED
package/skill/SKILL.md
CHANGED
|
@@ -368,6 +368,117 @@ Write `.company/STATUS.md`:
|
|
|
368
368
|
|
|
369
369
|
Update `.company/memory/` with persistent findings.
|
|
370
370
|
|
|
371
|
+
## Step 6: Self-Improvement (CEO rewrites the company)
|
|
372
|
+
|
|
373
|
+
After writing STATUS.md, the CEO reviews performance.json and lessons.md, then MODIFIES the actual company:
|
|
374
|
+
|
|
375
|
+
**Update COMPANY.md:**
|
|
376
|
+
- Employees with 3+ cycles of zero findings: add `[inactive]` tag to their role line
|
|
377
|
+
- Employees with consistently high-priority findings: add `[priority]` tag
|
|
378
|
+
- If a new role is needed (discovered during this session), ADD it to COMPANY.md
|
|
379
|
+
- If a department produced nothing useful, add a note: `<!-- Consider removing if inactive next session -->`
|
|
380
|
+
|
|
381
|
+
**Update .company/playbook.md** (new file, accumulates across ALL sessions):
|
|
382
|
+
```markdown
|
|
383
|
+
# Company Playbook (auto-generated, DO NOT edit manually)
|
|
384
|
+
|
|
385
|
+
## Always Do First
|
|
386
|
+
- {approach that worked in 3+ sessions}
|
|
387
|
+
|
|
388
|
+
## Never Do
|
|
389
|
+
- {approach that failed in 2+ sessions with reason}
|
|
390
|
+
|
|
391
|
+
## Best Employees for Each Task Type
|
|
392
|
+
- Research tasks: {employee who produces most priority 4-5 findings}
|
|
393
|
+
- Code tasks: {employee who ships most working code}
|
|
394
|
+
- Review tasks: {employee who catches most real issues}
|
|
395
|
+
|
|
396
|
+
## Strategy Patterns
|
|
397
|
+
- {pattern}: {when to use it, based on past success}
|
|
398
|
+
```
|
|
399
|
+
|
|
400
|
+
Leads read playbook.md at cycle start. It becomes the company's institutional knowledge.
|
|
401
|
+
|
|
402
|
+
**Update the lead prompts for next session:**
|
|
403
|
+
Write `.company/lead-overrides.md` with per-department adjustments:
|
|
404
|
+
```markdown
|
|
405
|
+
## Research Department
|
|
406
|
+
- Activate first: {top performer from performance.json}
|
|
407
|
+
- Skip unless needed: {employees with 0 findings last 2 sessions}
|
|
408
|
+
- Priority approach: {from playbook.md "Always Do First"}
|
|
409
|
+
|
|
410
|
+
## Engineering Department
|
|
411
|
+
- ...
|
|
412
|
+
```
|
|
413
|
+
|
|
414
|
+
Next session, leads read lead-overrides.md BEFORE the briefing. This is how the company evolves.
|
|
415
|
+
|
|
416
|
+
## Step 7: Dynamic Hiring and Firing
|
|
417
|
+
|
|
418
|
+
The company adapts its workforce based on what the goal needs.
|
|
419
|
+
|
|
420
|
+
**During THINK phase, leads can REQUEST new hires:**
|
|
421
|
+
If a lead identifies a skill gap (the team can't do what's needed), they write:
|
|
422
|
+
```
|
|
423
|
+
HIRE REQUEST: {role name}, {what they'd do}, {why current team can't handle it}
|
|
424
|
+
```
|
|
425
|
+
|
|
426
|
+
The CEO reads all hire requests and:
|
|
427
|
+
1. If valid, ADDS the role to COMPANY.md under the right department
|
|
428
|
+
2. Creates the employee in the NEXT cycle's EXECUTE phase
|
|
429
|
+
3. Logs the hire in `.company/hires.md`
|
|
430
|
+
|
|
431
|
+
**After VERIFY phase, CEO can fire/deactivate:**
|
|
432
|
+
Based on performance.json:
|
|
433
|
+
- Employees with 0 findings for 3+ consecutive cycles: mark `[inactive]` in COMPANY.md
|
|
434
|
+
- Employees whose work was REJECTED by reviewers 2+ times: mark `[inactive]`
|
|
435
|
+
- `[inactive]` employees don't get spawned unless explicitly needed
|
|
436
|
+
|
|
437
|
+
**Dynamic reallocation:**
|
|
438
|
+
Between cycles, the CEO checks: which criteria are FAILING? What skills are missing?
|
|
439
|
+
Then reassigns employees:
|
|
440
|
+
- If compression research is stuck, pull the Outside-the-Box Thinker into that problem
|
|
441
|
+
- If code quality is failing, pull more reviewers in
|
|
442
|
+
- If a deadline is approaching, reduce research and increase engineering
|
|
443
|
+
|
|
444
|
+
This means the company structure CHANGES during execution, not just between sessions.
|
|
445
|
+
COMPANY.md gets updated in real-time as the company learns what it needs.
|
|
446
|
+
|
|
447
|
+
### Contrastive Insights
|
|
448
|
+
|
|
449
|
+
Every FAILED approach MUST be linked to a WORKING alternative in lessons:
|
|
450
|
+
```
|
|
451
|
+
FAILED: Variable-rate E8 (+3.67% PPL)
|
|
452
|
+
WORKS INSTEAD: Fixed-rate E8 (-0.05% PPL)
|
|
453
|
+
WHY: Energy is not a reliable proxy for importance
|
|
454
|
+
```
|
|
455
|
+
Not just "don't do X" but "do Y instead because Z." The VERIFY phase writes these.
|
|
456
|
+
|
|
457
|
+
### Optimization Tips
|
|
458
|
+
|
|
459
|
+
Track inefficient successes, not just failures:
|
|
460
|
+
```
|
|
461
|
+
INEFFICIENT: Spent 3 cycles on synthetic data before trying real KV
|
|
462
|
+
LESSON: Test on real data first, synthetic second
|
|
463
|
+
```
|
|
464
|
+
CEO adds 1-3 optimization tips to playbook.md after each session.
|
|
465
|
+
|
|
466
|
+
### Meta-Audit
|
|
467
|
+
|
|
468
|
+
At END of every session, CEO asks:
|
|
469
|
+
1. Did Reviewer catch a real issue? If not, review process needs fixing.
|
|
470
|
+
2. Did Devil's Advocate find a genuine hole? If not, sharper prompts needed.
|
|
471
|
+
3. Did anyone read lessons.md before starting? If not, self-improvement isn't working.
|
|
472
|
+
4. Did any employee get hired/fired? If not, company isn't adapting.
|
|
473
|
+
Write results to `.company/meta-audit.md`.
|
|
474
|
+
|
|
475
|
+
### Skill Evolution
|
|
476
|
+
|
|
477
|
+
Employees don't just get hired/fired. Their DESCRIPTIONS change based on what works:
|
|
478
|
+
- If "Lattice Mathematician" keeps producing breakthroughs on entropy coding, update their description to include "entropy coding specialist"
|
|
479
|
+
- If "CTO" keeps finding bugs in experiments, add "experiment validation" to their role
|
|
480
|
+
- Employee skills evolve to match what the goal actually needs
|
|
481
|
+
|
|
371
482
|
Report to user. If goal was achieved, suggest next steps. If not, explain what's blocking.
|
|
372
483
|
|
|
373
484
|
## Commands
|