Agent Operations Hub

Single map for agents working in or against this wiki. After identity files, read this page once per session so you know what already applies, where the full tables live, and what to open before code, research, or wiki writes. Everything else is detail linked from here.

How to use this page

This hub is the routing map, not the full manual. Use it to answer four questions quickly:

What identity and maintenance context must be loaded?
Which rule layer already applies?
Which deeper resolver or skill should own the task?
What must be updated if this task changes agent behavior?

Do not read every linked page by default. Read the startup trio, this hub, the freshness report, then load only the protocol or skill that matches the work. The point is fast correct routing, not context hoarding.

Agent operating model

Kevin's agent system is built around a read/write loop:

Principle	Concrete behavior
Brain-first	Search `qmd` and the wiki before answering from memory or web.
Skills-first	Read the relevant installed skill before hand-rolling a workflow.
Graph-first writes	Wiki edits update pages, routing, links, index, qmd, and log.
No one-off work	Repeated work becomes a skill, script, or automation.
Proof loop	Work returns artifacts/traces that update memory, routing, skills, tests, or logs.
User sovereignty	Recommend clearly; Kevin decides contested or strategic calls.
Durable config	Agent behavior lives in `skills/`, `config/`, `AGENTS.md`, and meta pages, not scattered chat memory.
Progressive disclosure	Root `AGENTS.md` / `CLAUDE.md` / `CODEX.md` / `CURSOR.md` files are pointer indexes; detailed behavior belongs in linked skills, wiki pages, scripts, and generated command indexes.

This is why the hub points to both knowledge pages and executable surfaces. Good agents do not just know the rule; they load the place where the rule is enforced.

0. Source of truth naming

Treat this open wiki checkout as authoritative.

Everywhere agents describe paths or onboarding steps, use kevin-wiki - e.g. ~/Documents/GitHub/kevin-wiki, wiki/, config/, scripts/ (as in AGENTS.md § Source of truth naming). Ignore incidental clone directory suffixes such as my-wiki-real in instructional copy.

1. Read order (canonical session boot)

Step	Artifact	Purpose
1	`wiki/SOUL.md`, `wiki/USER.md`, `wiki/HEARTBEAT.md`	Identity, user model, operational cadence
2	This file (`wiki/meta/agent-operations-hub.md`)	Routing + enforcement map (you are here)
3	`npx tsx scripts/check-freshness.ts`	Overdue automations, stale syncs, maintenance flags
4	§ 5–§ 6 (workspace switch + task routing)	Load depth only when needed

Source: AGENTS.md Session Startup + this hub, 2026-05-11

2. Three layers of “what applies”

Layer	What it is	Where
A. Always-on rules	Brief universal rules injected every message (~3.3K tok): brain-first, quality, test-before-bulk, no-one-off, ripgrep, process-hygiene, agent-docs-mesh, signal-detector, wiki-source-of-truth. Heavier / situational rules are on-demand (§ 3).	`config/cursor/rules/.mdc` (→ `~/.cursor/rules`, all repos) + `.cursor/rules/.mdc` (wiki repo only)
B. Deep dispatcher	Full service registry, task categories, Dedalus inventory, design-review rows	`wiki/SKILL-RESOLVER.md` (on-demand)
C. Capability router	Generic ranked interface per service + tool per task. Harness-specific files mirror this for local loader constraints.	`wiki/meta/capability-routing-map.md` (canonical) + `wiki/SKILL-RESOLVER.md`; Cursor projection: `config/cursor/rules/tool-hierarchy.mdc`

Rule: If the generic wiki pages and a harness projection disagree, the wiki wins. Update Capability Routing Map and Skill Resolver first, then mirror the compact shape into the relevant harness file. tool-hierarchy.mdc is Cursor's pocket summary, not the canonical agent brain.

Source: compiled from repo layout, 2026-05-11

3. Rules — always-on vs fetched on demand

A few brief, universal rules are alwaysApply (every message, ~3.3K tok total). The heavier / situational rules are Agent-Requested (description + alwaysApply:false — the agent fetches them when the description matches the task) or Auto-Attached (load only when matching files are in context). Content is identical either way — this keeps the harness lean on every project (Cursor docs: reserve alwaysApply for a few brief rules). Missing locally? Run npm run bootstrap.

Always-on (every message):

Rule	What you must not forget
`wiki-source-of-truth.mdc`	Wiki paths, `_index`, SKILL-RESOLVER pointer
`conventions-brain-first.mdc`	`qmd search` / `qmd query` before web/API
`conventions-quality.mdc`	Wiki writes: citations, backlinks, compiled vs timeline
`conventions-test-before-bulk.mdc`	Batch work: sample 3–5, approve, then bulk
`no-one-off-work.mdc`	Repeatable work → skill or automation
`ripgrep.mdc`	Use `rg`, never `grep`
`process-hygiene.mdc`	Own every process you spawn; tear down before ending a turn
`agent-docs-mesh.mdc`	Per-repo AGENTS.md + nearest SKILL.md; enroll on entry
`signal-detector.mdc`	Capture ideas + entities after substantive replies

On-demand — fetched when relevant (don't assume it's in context; the description triggers it):

Rule	Fires when
`tool-hierarchy.mdc`	Cursor-only compact projection for choosing any tool / service / CLI / MCP / skill; canonical routing remains Capability Routing Map + Skill Resolver
`agent-native-cli.mdc`	Accessing an external API / app / site
`pr-review-lessons.mdc`	Reviewing or shipping code
`css-ui-enforcement.mdc`	Auto-attaches on `*.{tsx,jsx,css,scss}`
`last30days-research.mdc`	"What's happening with X" / research
`agent-reach.mdc`	Internet / live-data tasks
`ideation-paradigms.mdc`	Ideas / startup-strategy questions

Source: rule-typing pass 2026-06-13; .cursor/rules/ + config/cursor/rules/

4. Where skills and config live

Kind	Path	Notes
Category skill source	`skills/{engineering,productivity,personal,misc,in-progress,deprecated}/<slug>/SKILL.md`	Canonical executable source tree
Runtime skill index	`skills/.runtime/all/<slug>`	Generated flat symlink index for Cursor, Claude Code, Codex, and AgentSkills
System skills	`skills/.system/<slug>/SKILL.md`	Hidden system skills projected into runtime index
Project AgentSkills links	`.agents/skills/<slug>`	Discoverability layer; prefer links into `skills/.runtime/all`
Dedalus rules/hooks reference	`config/dedalus/agents/`	Rules, hooks, subagents; no skill bodies
Dedalus Cursor product rules	`config/dedalus/cursor-rules/*.mdc`	Copy into monorepo when refreshing
Skill registry + family synthesis	`wiki/meta/skill-registry.{json,md}` + `wiki/concepts/*-skills.md` + `wiki/skills/<category>/`	Generated executable inventory, readable family pages, and generated source views; no hand-written root skill mirrors
Automations	`automations/<slug>.md`	Cron / manual runs

Historical Dedalus reference tables: **wiki/SKILL-RESOLVER.md § Dedalus monorepo**. Use only for an explicit legacy task; Kevin's affiliation ended 2026-07-06. Source: User request, 2026-07-14

Source: AGENTS.md “Where Things Live”, config-sync, 2026-05-11

5. Workspace switch (which forest first?)

You are in…	Load first
This wiki repo (`kevin-wiki`)	SKILL-RESOLVER (personal + operational sections), brain-first, quality
Explicit historical Dedalus task	SKILL-RESOLVER § Dedalus monorepo + preserved `config/dedalus/agents/rules/`; do not assume current access, employment, or authority
Another repo with Kevin	`agent-docs` skill + root `AGENTS.md` + nearest `SKILL.md`; Capability Routing Map for tool choice; harness projection only if the active harness needs it; still brain-first if the question is about people/tools/concepts in the wiki

Source: User directive, 2026-05-11

6. Task → first action (routing)

I’m about to…	Open / run first
Answer from compiled Kevin knowledge	The Brain-Agent Loop - `qmd search` / `qmd query` / `qmd get` before web
Resolve a known old slug after wiki compaction	`npm run wiki:resolve -- <slug-or-title>`; then `qmd get` the returned canonical page
Pick tool, MCP, or CLI	Capability Routing Map → `wiki/SKILL-RESOLVER.md` if non-obvious → harness projection only for local syntax/loader details
Pick between overlapping tool families	Capability Routing Map — browser/computer use, live search vs wiki search, model choice, service access
File, create, merge, or recategorize wiki pages	Wiki Brain Operating Model first for shape/merge rules, then `wiki/RESOLVER.md` for filing + `conventions-quality.mdc`
Browse topics	`wiki/_index.md`
List slash commands	Slash Command Index - generated from skill sources and `package.json`; aliases `/to-spec` and `/to-tickets` route to Kevin's existing owners
Draft a post / tweet / LinkedIn	`social-draft` skill (1st) → humanizer + kevin-voice polish before publish; full ranking in SKILL-RESOLVER § Career and Content
Full schema / ingest / operations	`AGENTS.md` (repo root)
Onboarding / repo overview drift	Root `README.md` - narrative must match reality; run `build-index` for stats table
UI: React, Tailwind, CSS	`wiki/style/css-ui-enforcement.md`, `wiki/style/react.md`
Bulk ingest or 10+ files	`conventions-test-before-bulk.mdc`
Kevin says add it / pastes URL, X, repo	`[[capture-ingest-protocol]]` - full wiki + routing + index + log (never one file only)
Edit or discuss `wiki/career/content-backlog.md`	`[[content-backlog]]` - end of turn: remind Kevin which rows are not `Posted` (see that page § Agent protocol); for `Ready to post`, include humanizer + kevin-voice / social-draft before publish
Live “what’s new” research	`last30days` skill + wiki
Access an external API / app / site	`agent-native-cli.mdc` → PrintingPress `pp-cli` (existing or `/printing-press`-printed); Agent Cookie for headless/second-Mac auth (Agent-Native CLIs)
Move / ingest data (DB ↔ warehouse, SaaS → warehouse)	Ingestr - One-Command Data Movement for Agents — `ingestr ingest --source-uri … --dest-uri …`, incremental loading; over hand-rolled ETL
Understand what an agent just built (stay in the loop)	`teacher-mode` skill (`/teacher`, "teach me", "make sure I understand") — incremental checklist, restate-first, quiz, completion gate (Staying in the Loop with Agents)
Plan / stress-test an app, feature, or refactor before PRD/code	`grill-with-docs` skill — map the design tree breadth-first, then walk load-bearing branches
Ship AI output / build an AI feature without slop	The Eval Loop (Slop Is an Output Problem) — score every output 0–1 vs a benchmark, gate < 0.7, failures → new test cases; runs in Hermes Agent (Nous Research)
Benchmark or self-improve an agent	`agent-eval-library` → Agent Self-Improvement Eval Library; emit a receipt, run `npm run agent-self-eval`, harden the evaluator via Red Queen Gödel Machine
Work on Claude Code / Hermes / OpenClaw / Codex	Agent Harness Capsules → load the runtime capsule (`claude-code-agent-stack`, `hermes-agent-stack`, `openclaw-agent-stack`, or `codex-agent-stack`) plus its canonical harness page
Work on Kevin's Hermes Mac Mini operator stack	`hermes-operator-stack` → Hermes Mac Mini Agent OS; then Hindsight Memory Provider, Honcho, Browser Testing Skills, and Agent Client Protocol for memory, browser, and coding-worker boundaries
Craft or audit a repeatable agent loop	`loopy` → Loopy; require goal, check, next action, stop condition, and handoff condition before codifying with `skill-creator`
Create or edit documents, spreadsheets, PDFs, decks, or templates	`documents` / `spreadsheets` / `pdf` / `presentations` / `template-creator` imported Codex skills; preserve bundled scripts, references, assets, templates, and `agents/openai.yaml`
Build, save, version, or deploy an OpenAI Sites artifact	`sites-building` then `sites-hosting`; read `.openai/hosting.json`, reuse project IDs, save source state before deploying
Use Cursor-specific onboarding, automation, or review subagents	`onboard`, `automate`, `review`, `review-bugbot`, `review-security`; only when the active harness supports the Cursor feature
Drain intelligence into plans (plan now with the strongest model, execute later cheap)	Borrowed Intelligence (Drain It into Plans) → Frontend and Design Skills (`/improve`, installed) → repo-root `plans/` catalog; execute via Browser Testing Skills
"What should I be using" / recommend skills+loops	Active Stack (What to Actually Use) — curated by workflow; every capture must land there + routing (Capture Ingest Protocol Activation gate)
Ship agent config on a machine	`wiki/meta/config-sync.md`, `npm run bootstrap`
Track what local processes are running / Agent Broom	`process-hygiene.mdc` -> `agent-hygiene.sh` via `cleanup-terminals-browsers`; Agent Broom is the portable package form; run `agent-hygiene.sh audit` before starting, during handoff, or when asked "what is running" (Agent Process Tracking, Audit, And Cleanup)
High CPU / clean up terminals & browsers	`process-hygiene.mdc` -> `agent-hygiene.sh stop
End a turn that spawned long-runners	Run `agent-hygiene.sh audit`; report what remains running and tear down only non-essential dev servers / hung test watchers (Agent Process Tracking, Audit, And Cleanup)
Run recurring job	`automations/` + `state.json` / freshness script
Refresh capability discovery	Capability Harvest Pattern → `npm run harvest:capabilities` → `npm run promote:capabilities` → registry + pattern/source snapshots, audited routing, Service CLI Coverage
Use newest promoted skills	Active Stack (What to Actually Use) + Skill Resolver service rows → installed skill in the correct category folder (source locked in `skills-lock.json`)
Audit wiki article/tool/skill quality	Wiki Quality Audit → `npm run wiki-quality` (add `-- --network --limit=N` for sampled live source checks), then fix read-first docs, hubs, tweet/artifact evidence, and bloated skills
Audit every page for pack/capsule routing	Agent Pack Corpus Audit → `npm run agent-packs:audit`; regenerate review-queue-fed packs with `npm run agent-packs:refresh`
Audit X bookmark artifacts one by one	X Bookmark Artifact Audit → `npm run audit:x-bookmark-artifacts` → `npx tsx scripts/doctor.ts --only x-bookmarks`; inspect enriched records, links, threads, replies, images, video demos, then record article/skill/tool/capsule/version decisions and promote thoughts into pages/skills/tools/playbooks/diagrams/capsules
Audit pages that should be merged, folded, or routed together	Page Consolidation Audit → `npm run wiki-consolidation:audit`; regenerate consolidation-queue-fed packs with `npm run agent-packs:refresh`
Heal wiki drift / close the doctor loop	`npm run self-heal` (triage + selective digest) → `npm run self-heal -- --apply` (safe deterministic fixes + verify); content fixes + PR via `automations/self-heal.md` (Self-Maintaining Systems)
Enter / enroll a Kevin project	`agent-docs` skill — check `.agent-docs/` + markers; `init` → fill → `doctor`; registry at `config/agent-docs/registry.json` (Agent-Docs Mesh)
Refresh project agent docs	`agent-docs scaffold` + `doctor --json` after substantive edits (Agent-Docs Mesh)
Create / open-source a public repo	`ensure-repo-license` skill — add/audit MIT license before publishing

Source: synthesized from rules + AGENTS.md, 2026-05-11; social backlog gate, 2026-05-13

7. Foundational Rules (non-negotiable)

Every agent, every session, every task:

Think Before Coding - State assumptions explicitly. Ask rather than guess. Push back when a simpler approach exists. Stop when confused.
Simplicity First - Minimum code that solves the problem. Nothing speculative. No abstractions for single-use code.
Surgical Changes - Touch only what you must. Don't improve adjacent code. Match existing style. Don't refactor what isn't broken.
Goal-Driven Execution - Define success criteria. Loop until verified. Strong success criteria let Claude loop independently.

Full rationale and hard limits: Agent Ethos.

Capture Ingest Protocol - mandatory multi-surface integration when Kevin pastes a link/repo/post and says add it
Agent Company OS - workspace brain + memory/skill overlay + command plane + proof ledger model for self-improving agent organizations
Capability Routing Map - capability-family hierarchy; explains intentional skill/tool mirrors
The Brain-Agent Loop - read wiki before answer, write after learn
The Resolver Pattern - dispatcher + specialized handlers
Agent Ethos - minimal change, hard limits, foundational rules
No One-Off Work - codify repeats
Wiki Quality Audit - broad audit for read-first docs, routers, source-backed pages, tweet/artifact evidence, and skill prompt load
Brain Capsules - public capsule interface that exposes source pages, skills, tools, packs, and maintenance rules
X Bookmark Artifact Audit - one-by-one X bookmark source, artifact, image, video, and capsule-promotion ledger
Agent-Docs Mesh - per-repo AGENTS.md + SKILL.md mesh, docs-doctor, enrollment registry
Hermes Mac Mini Agent OS - local persistent Hermes operator stack: Mac Mini, Hindsight, Honcho, qmd/Obsidian, agent-browser, ACP/SSH workers, and launchd/no-agent monitors
Cross-Agent Harness Portability - one harness for Cursor / Claude Code / Codex / OpenClaw / Hermes: skill dirs, identity-file mapping, bootstrap vs remote

Timeline

2026-07-06 | Added the progressive-disclosure doctrine for agent entry files: root/provider AGENTS.md, CLAUDE.md, CODEX.md, CURSOR.md, and .cursorrules should be pointer indexes, while detailed behavior lives in skills, wiki meta pages, scripts, and the generated Slash Command Index. Source: User request and screenshot, 2026-07-06
2026-07-04 | Added the Hermes Mac Mini operator route: use hermes-operator-stack and Hermes Mac Mini Agent OS before changing Hindsight, Honcho, agent-browser, ACP/SSH worker, Telegram, launchd, or Hermes Desktop configuration. Source: User request, 2026-07-04
2026-06-29 | Clarified that Capability Routing Map and Skill Resolver are the generic agent-brain source of truth for tool routing; config/cursor/rules/tool-hierarchy.mdc is only the Cursor projection. Source: User correction, 2026-06-29
2026-06-29 | Added brain capsules and X bookmark artifact audit routing after Kevin requested capsules as the public brain interface and a literal one-by-one review of 800+ bookmark artifacts/images. Source: User request, 2026-06-29
2026-06-29 | Added self-improvement eval and Loopy loop-skill routes to the task matrix. Source: User request, 2026-06-29; Agent Self-Improvement Eval Library; Loopy
2026-06-28 | Updated skill routing paths from runtime namespaces to category source folders plus generated skills/.runtime/all, with Dedalus-origin skills folded into the shared taxonomy. Source: User request, 2026-06-28
2026-06-26 | Tightened process-hygiene routing for localhost sprawl: agents must track/audit local processes with the process ledger and listener audit before starting another dev server, then report or stop recorded long-runners at end of turn. Source: User request and correction, 2026-06-26
2026-06-25 | Added wiki-quality-audit as the first route for broad wiki/article/tool/skill quality passes, including optional live URL sampling. Source: User request, 2026-06-25; scripts/wiki-quality-audit.ts
2026-06-25 | Added Agent Company OS to the operating model: agent work should return proof artifacts/traces that update memory, routing, skills, tests, or logs. Source: X/@DerekNee, 2026-06-25
2026-06-25 | Expanded capability-harvest routing beyond skills/MCP/CLI registries to include pattern/source snapshots for templates, design techniques, browser/testing loops, eval/review/security methods, data workflows, runtime substrates, and automations. Source: User request, 2026-06-25
2026-06-22 | Extended capability-harvest routing with npm run promote:capabilities and Service CLI Coverage so broad registry harvests turn into audited promotion candidates. Source: User request, 2026-06-22
2026-06-22 | Added the capability-harvest route for broad skills.sh/MCP/CLI registry refreshes. Source: User request, 2026-06-22
2026-06-18 | Added explicit usage guidance and the agent operating model so the hub explains how to route with the linked protocols instead of acting as a flat link catalog. Source: User request, 2026-06-18
2026-06-16 | Added the self-heal loop (scripts/self-heal.ts + automations/self-heal.md): triage, per-finding stand-down state, verified auto-safe fixes, selective notification — closing the doctor's detection into healing (Self-Maintaining Systems). De-noised check-freshness (cloud automations → notice, .cursor-automation.md excluded). § 6 routing row added. Source: User, 2026-06-16
2026-06-17 | Added explicit first-action routes for grill-with-docs planning and ensure-repo-license repo publication checks. Source: section audit, 2026-06-17
2026-06-17 | Added Capability Routing Map as the route for overlapping tool families and intentional skill/tool mirror semantics. Source: User, 2026-06-17
2026-06-07 | Agent-Docs Mesh codified: agent-docs skill, agent-docs-mesh.mdc always-on rule, hub § 5–6 rows, AGENTS.md Layer 10. Source: User, 2026-06-07
2026-06-05 | Added always-on process-hygiene.mdc + cleanup-terminals-browsers skill: own/teardown spawned processes, kill stray dev/test/browser by pgid, keep IDE/MCP. Hub § 3 + § 6 rows. Source: User, 2026-06-05
2026-05-13 | Content-backlog task row: add humanizer + kevin-voice / social-draft for Ready to post rows. Source: User, 2026-05-13
2026-05-12 | Task matrix: row for Content Backlog edit → pending-publish reminder protocol. Source: User, 2026-05-12
2026-05-11 | README maintenance: hub task row for onboarding doc drift; build-index syncs README stats table. Source: User, 2026-05-11
2026-05-11 | Linked Capture Ingest Protocol for chat captures; fixed frontmatter tier. Source: User, 2026-05-11
2026-05-11 | Initial hub: unified read order, always-on checklist, skill locations, workspace switch, task matrix, links to SKILL-RESOLVER / AGENTS / RESOLVER. Source: User, 2026-05-11

Source: AGENTS.md Session Startup + this hub, 2026-05-11

Source: compiled from repo layout, 2026-05-11

Source: rule-typing pass 2026-06-13; .cursor/rules/ + config/cursor/rules/

Source: User request, 2026-07-14

Source: X/@DerekNee, 2026-06-25

Source: section audit, 2026-06-17

Agent Operations Hub

How to use this page

Agent operating model

0. Source of truth naming

1. Read order (canonical session boot)

2. Three layers of “what applies”

3. Rules — always-on vs fetched on demand

4. Where skills and config live

5. Workspace switch (which forest first?)

6. Task → first action (routing)

7. Foundational Rules (non-negotiable)

Timeline

Related Pages

44 pages link here

Agent Operations Hub

How to use this page

Agent operating model

0. Source of truth naming

1. Read order (canonical session boot)

2. Three layers of “what applies”

3. Rules — always-on vs fetched on demand

4. Where skills and config live

5. Workspace switch (which forest first?)

6. Task → first action (routing)

7. Foundational Rules (non-negotiable)

8. Related mental models

Timeline

Related Pages

44 pages link here