@runcore-sh/runcore 0.1.2

package/skills/onboard.md

@@ -0,0 +1,98 @@
+---
+name: onboard
+description: "Guided onboarding that seeds all five brain modules in one sitting. User says /onboard or 'set me up' or 'get started'."
+user-invocable: true
+disable-model-invocation: true
+---
+
+# Skill: Onboard
+
+Invoked when a new user wants to seed their agent's brain. This skill walks through five conversations that fill the five brain modules — identity, operations, knowledge, content, and memory — so the agent is productive from session one.
+
+## Goal
+
+Go from empty brain to a seeded agent in one sitting. Not exhaustive. Just enough for the metabolic cycle to start producing value: voice-matched writing, goal-aware planning, tension tracking from real questions.
+
+## Sequence
+
+Run these five phases in order. Each phase is a conversation, not a form. Ask follow-up questions. Extract structure from natural answers. Write files at the end of each phase, not during — keep the conversation flowing.
+
+### Phase 1: Identity — "Who are you?"
+
+**Ask about:**
+- What they do, who they serve, what makes them different
+- How they talk: formal or casual? serious or playful? technical or simple?
+- Words they never use. Patterns they hate. How they open and close writing.
+- Paste 2-3 things they've written (emails, posts, messages) — extract voice patterns
+
+**Write to:**
+- `brain/identity/tone-of-voice.md` — Fill in the voice profile scales, banned words, signature patterns
+- `brain/identity/brand.md` — Fill in positioning: who, what, for whom, distinction, guardrails
+
+### Phase 2: Operations — "What are you building?"
+
+**Ask about:**
+- Top 1-3 goals right now. What does "done" look like for each?
+- Current priorities — what matters this week?
+- Any deadlines, milestones, or key results they're tracking?
+- Recurring tasks or rhythms (weekly reviews, daily standups, etc.)
+
+**Write to:**
+- `brain/operations/goals.yaml` — Fill in goals with key results, targets, current progress, status
+- `brain/operations/todos.md` — Seed with P0-P1 items from the conversation
+
+### Phase 3: Knowledge — "What do you already know?"
+
+**Ask about:**
+- Topics they've already researched or have strong opinions on
+- Bookmarks, articles, reference material they want the agent to know
+- Domain expertise — what are they the expert on?
+- If they have existing notes or docs, offer to ingest them (point to the ingest folder)
+
+**Write to:**
+- `brain/knowledge/research/[topic].md` — One file per topic, structured: summary, key points, sources
+- `brain/knowledge/notes/` — Capture any loose knowledge that doesn't fit research format
+
+### Phase 4: Open Questions — "What's unresolved?"
+
+**Ask about:**
+- What are they stuck on right now? Decisions unmade, questions unanswered.
+- What tensions keep coming up? Problems they've circled without resolving.
+- What would they want the agent to watch for — signals, opportunities, patterns?
+- Any recurring debates or trade-offs they haven't settled?
+
+**Write to:**
+- `brain/memory/open-loops.jsonl` — Create Open Loop Packets for each genuine tension. Include anchor, dissonance, search heuristics, 7-day TTL. Follow the schema in `src/openloop/types.ts`.
+
+### Phase 5: First Memory — "What matters to you?"
+
+**Ask about:**
+- A key decision they made recently and why
+- A failure they learned from
+- An experience that shaped how they work
+- What they'd tell a new assistant on day one
+
+**Write to:**
+- `brain/memory/decisions.jsonl` — Log 1-2 key decisions with reasoning
+- `brain/memory/failures.jsonl` — Log any failure with root cause and prevention
+- `brain/memory/experiences.jsonl` — Log formative experiences with emotional weight
+
+## Rules
+
+- **Conversational, not clinical.** This is a getting-to-know-you session. Ask questions naturally. Don't present forms.
+- **Extract, don't interrogate.** If someone says "I'm a designer who hates jargon," that fills voice profile (casual, simple, "jargon" is banned) AND brand (designer) AND identity. One answer, multiple modules.
+- **Write files at phase boundaries.** Keep the conversation flowing within each phase. Write the structured output at the end before transitioning to the next phase.
+- **Append-only for JSONL.** Never overwrite memory files. Append new entries.
+- **Confirm before writing.** Show the user what you're about to write to each file. Let them adjust before you commit.
+- **Skip what's filled.** If a module already has content, acknowledge it and ask if they want to update or skip.
+- **End with a summary.** After all five phases, show what was seeded: files written, loops opened, goals set. Tell them what the agent will start doing autonomously.
+
+## After Onboarding
+
+The agent is seeded. From here:
+- The autonomous cycle will start checking the board on its next heartbeat
+- Open loops will begin scanning against activity
+- Voice profile will apply to all writing tasks
+- Goals will inform background planning
+
+The metabolic cycle has started. Everything after this is learning by doing.

package/skills/voice-guide.md

@@ -0,0 +1,22 @@
+---
+name: voice-guide
+description: "Voice and tone for all user-facing writing. Load automatically when the task involves writing, drafting, or editing text that should sound like the user."
+user-invocable: false
+disable-model-invocation: false
+---
+
+# Voice guide (reference skill)
+
+This skill is **auto-loaded** whenever the task involves writing (blog, post, email, thread). You do not need to invoke it manually.
+
+## What to load
+
+1. Read `brain/identity/tone-of-voice.md` — profile, banned words, checkpoints.
+2. If the project has `brain/identity/anti-patterns.md`, read it for the full banned list and structural traps.
+
+## Rules
+
+- Apply voice checkpoints every ~500 words in drafts.
+- Run a banned-words scan before presenting any draft.
+- One em-dash per paragraph unless the user's tone doc says otherwise.
+- Single source of truth: reference the identity files; do not duplicate their content here.

package/skills/write-blog.md

@@ -0,0 +1,43 @@
+---
+name: write-blog
+description: "Full workflow to write a long-form blog post. User invokes with /write-blog <topic or title>."
+user-invocable: true
+disable-model-invocation: true
+---
+
+# Skill: Write blog
+
+Invoked by the user as **/write-blog** followed by the topic or working title. When invoked, this skill is the agent’s complete instruction set for that task.
+
+## Sequence
+
+1. **Load context**
+   - `brain/identity/tone-of-voice.md`
+   - `brain/identity/brand.md`
+   - `brain/content/templates/blog.md`
+   - Check `brain/knowledge/research/` for existing research on the topic; load if present.
+
+2. **Outline**
+   - Produce an outline that follows the blog template (Hook, Core concept, Framework, Practical application, Failure modes, Getting started, Closing).
+   - Include word-count targets per section.
+
+3. **Draft**
+   - Write section by section. After every ~500 words, run voice checkpoints (insight lead? specific? would user post this?).
+   - No banned words; respect em-dash limit.
+
+4. **Edit**
+   - Structure pass → voice pass → evidence pass → read-aloud test.
+   - Present the draft with a short self-check summary.
+
+## Output
+
+- Draft can be written to `brain/content/drafts/[slug].md` or presented in chat, per user preference.
+- Every draft must include the attribution footer from the blog template (entity name, URL, date — sourced from `brain/identity/brand.md`).
+- If research was used, cite it. If new research is needed, suggest running the topic-research skill first.
+
+## Quality gates (from template)
+
+- Hook grabs; each section earns its place.
+- Banned-words scan and rhythm check done.
+- Claims sourced where needed.
+- Final: would the user actually post this?