Capabilities & Infrastructure

Procedural manuals, distilled heuristics, host-specific reality, and dynamic ops guides.

SKILL 🎪 Procedural Manual

Actionable API/tool logic utilizing progressive disclosure (reads YAML frontmatter, loads body on-demand).

Skill Anatomy

~/.hermes/skills/<category>/<skill-name>/
├── SKILL.md              # Full executable workflow (YAML frontmatter + markdown)
├── references/           # Supporting docs (API specs, patterns, pitfalls)
│   ├── api.md
│   ├── patterns.md
│   └── pitfalls.md
├── templates/            # Reusable templates
│   └── config.yaml.tmpl
├── scripts/              # Automation scripts
│   └── validate.py
└── assets/               # Static assets
    └── icon.png

Frontmatter Schema

---
name: "skill-name"
version: "1.0.0"
category: "devops|frontend|docs|mlops|..."
description: "One-line purpose"
triggers:
  - "user asks about X"
  - "task involves Y"
toolsets: ["terminal", "file", "web", "browser"]
pitfalls:
  - "Common mistake 1"
  - "Common mistake 2"
verification:
  - "How to confirm success"
---

Progressive Disclosure

Load Stage	Content	Token Cost
List	`skills_list()` — name + description only	~50
View	`skill_view(name)` — frontmatter + body	~2K
Reference	`skill_view(name, file_path='references/x.md')`	~1K

Skill Lifecycle

graph LR
    A[Create] --> B[Test]
    B --> C[Patch on issues]
    C --> D[Pin if critical]
    D --> E[Prune if stale]

Current Active Skills (7 with SKILL.md)

Skill	Category	Purpose
`hermes-essentials`	hermes-essentials	Core config, debugging, planning, TDD
`superpower-pipeline`	superpower-pipeline	Research + web search pipeline
`superpower-research`	superpower-research	Deep research automation
`omp-suite`	omp-suite	OMP plugin agent
`hugo-llm-wiki`	hugo-llm-wiki	Hugo docs site patterns (cron-loaded)
`thermo-nuclear-code-quality-review`	thermo-nuclear-code-quality-review	Strict maintainability review
`transcript-extract`	transcript-extract	Extract transcripts from URLs

PROCEDURES 🧠 Learned Dispositions

Distilled records of effective strategies and heuristic patterns extracted from past experiences.

Format

## Procedure: <name>
 
**Context**: When this applies
**Pattern**: The heuristic (if X, then Y)
**Evidence**: Sessions where this worked
**Confidence**: 0.0-1.0
**Last Validated**: YYYY-MM-DD

Current Procedures

Procedure	Pattern	Confidence
Skill-first	Always `skill_view` before tool use for known domains	0.95
Memory-first	Check `session_search` before web search for past context	0.90
Verify handles	Never trust subagent “success” — fetch URL, stat file, read back	1.00
Cron env	Always `source ~/.hermes/.env` in cron scripts	1.00
Fetch timeout	Always `withTimeout()` + safety timer for Svelte fetch	0.98
Patch skills	Update skill immediately when pitfall discovered	0.92

Extraction Protocol

Trigger — Repeated success (3+ times) or user correction
Distill — Write as heuristic, not instruction
Validate — Apply next 2 times, confirm
Record — Add to PROCEDURES with confidence score

TOOLS 🔌 Physical Reality

Host-specific connection strings, local IP addresses, DB paths, and timeout policies.

Connection Registry

Tool	Connection	Timeout	Notes
PostgreSQL	`postgresql://user:pass@localhost:5432/db`	30s	Via `~/.pgpass`
SQLite (kanban)	`~/.hermes/kanban.db`	10s	OMP Suite
SQLite (state)	`~/.hermes/state.db`	10s	Session/usage stats
Nginx	`systemctl reload nginx`	10s	Requires sudo
Hugo	`hugo --minify`	60s	From `~/hugo-llm-wiki/`
Discord Bot	`DISCORD_TOKEN` from `.env`	—	Never in execute_code
OpenRouter	`OPENROUTER_API_KEY` from `.env`	180s	Gateway timeout

Timeout Policies

timeouts:
  terminal_foreground: 180s
  terminal_background: 600s
  gateway_request: 1800s
  fetch_with_timeout: 30s
  skill_load: 10s
  memory_inject: 5s
  delegate_task: 300s

Path Mappings

Logical	Physical
`~/.hermes/config.yaml`	Main config
`~/.hermes/.env`	API keys (NOT inherited by crons)
`~/.hermes/skills/`	Installed skills
`~/.hermes/cron/`	Cron job output
`~/.hermes/logs/`	agent.log, errors.log, gateway.log
`~/.hermes/dreams/`	Dream Swarm reports
`~/.hermes/plugins/`	Hermes plugins (OMP Suite)
`~/hugo-llm-wiki/`	LLM Wiki source
`~/svelte-flexiboards-dashboard/`	Svelte 5 dashboard

checklists_dir ✅ Ops Guides

Step-by-step pre-flight and execution guides (checklists/*.md) referenced dynamically.

Structure

checklists/
├── pre-flight.md          # Before any major operation
├── deploy.md              # Hugo / Svelte / Nginx deploy
├── cron-create.md         # New cron job checklist
├── skill-create.md        # New skill checklist
├── subagent-spawn.md      # delegate_task checklist
├── memory-write.md        # memory tool checklist
├── model-switch.md        # Config model change
└── incident.md            # Production incident response

Checklist Format

# Deploy Checklist
 
## Pre-Conditions
- [ ] Working directory correct
- [ ] No uncommitted changes (or intentional)
- [ ] Tests pass locally
 
## Execution
- [ ] `hugo --minify` succeeds
- [ ] `nginx -t` passes
- [ ] `systemctl reload nginx` succeeds
 
## Verification
- [ ] `curl -s https://llm-wiki.0rk.de | grep -q "LLM Wiki"`
- [ ] JSON feed valid: `curl -s /topics/index.json | python3 -m json.tool`
 
## Rollback
- [ ] Previous public/ backed up
- [ ] Rollback command documented

Dynamic Reference

Skills and cron jobs reference checklists by name:

# In skill script
source ~/hugo-llm-wiki/checklists/deploy.md

Cross-References

Identity Anchors → LEXICON defines SKILL terminology
Orchestration → OPS gates TOOL connections
Autonomic → HEARTBEAT runs checklists on schedule
Spec-Driven → tasks maps to checklists_dir

Capabilities & Infrastructure is the executable layer — it turns architecture into running code.

🧭 Explore

Graph View