content-planner

Files (2)

• SKILL.md
• scan_repos.py · scripts/

SKILL.md (first lines)

# Content Planner

Generate a content plan from recent engineering work. Scan GitHub repos + buildlog entries, identify publishable narratives with fan-out opportunities, and output Obsidian-compatible markdown.

## Step 0: Load Conventions

Before doing ANYTHING, read the shared conventions file:

```
Read ${SKILLS_DIR}/_conventions.md
```

This file defines: canonical vault path, folder-to-type mapping, frontmatter contract, valid statuses, tag hierarchy, cross-reference syntax, and the PipelineEnvelope schema. All output from this skill MUST conform to those conventions. If there is a conflict between this SKILL.md and `_conventions.md`, the conventions file wins.

## Workflow

### Step 1: Scan Recent Activity + Buildlog

Run the Python scan script to collect GH data AND buildlog entries:

```bash
python3 ${SKILLS_DIR}/content-planner/scripts/scan_repos.py --days 14 --format summary
```

For JSON output (when you need structured data for processing):

```bash
python3 ${SKILLS_DIR}/content-planner/scripts/scan_repos.py --days 14 --format json
```

To scan specific repos only:

```bash
python3 ${SKILLS_DIR}/content-planner/scripts/scan_repos.py --days 14 --repos "/path/to/repo1,/path/to/repo2"
```

Key repos to always scan (in priority order):
1. `qortex-track-c` — core engine
2. `buildlog-template` — dev workflow tool
3. `openclaw` — AI agent platform
4. `interlinear` — language learning app
5. `MindMirror` — health/wellness
6. `langchain-qortex`, `mastra-qortex` — consumer packages

The scan script automatically collects:
- **GitHub**: merged PRs, open issues, recent commits
- **Buildlog entries**: title, sections, commits, word count, depth classification
- **Fan-out analysis**: which content types each entry can generate
- **Metadata**: review learnings, promoted skills, reward events

### Step 2: Read Deep Buildlog Entries

The scan script classifies entries by depth:
- **deep** (800+ words): Full narrative entries. Read these in full — they're gold mines.
- **medium** (300-800 words): Solid entries. Read selectively based on fan-out score.
- **shallow** (<300 words): Quick commit logs. Skim for context, don't build content from these alone.

For each **deep** and high-fanout **medium** entry, read the full file:

```
Read {entry.file}
```

When reading, extract:
1. **The narrative arc**: Goal → Obstacle → Resolution → Insight
2. **Concrete artifacts**: commit hashes, test counts, code snippets, error messages
3. **Quotable moments**: "what went wrong" stories, unexpected discoveries, aha moments
4. **Cross-references**: PR/issue links, other repos mentioned, buildlog entries that connect

### Step 3: Fan-Out Analysis

Each buildlog entry can generate **multiple** content pieces. The scan script tags fan-out opportunities automatically:

| Entry Section | Content Type | Channel | Example |
|---------------|-------------|---------|---------|
| The Goal / Context | **Provocation** | LinkedIn | "Nobody's doing X. Here's what that costs." |
| Architecture / Components | **Architecture** | Blog | "Here's why we chose X over Y." |
| The Journey / What Went Wrong | **Build Log** | Substack | "I shipped X in Y days. Here's how." |
| Improvements / Lessons | **Lessons** | LinkedIn | "I learned X the hard way. Save yourself the trip." |
| Test Results / Benchmarks | **Results** | Blog | "We measured X. The data says Y." |

**Fan-out scoring**:
- `fanout_count >= 4`: High-value entry. Plan 3-5 content pieces from it.
- `fanout_count 2-3`: Medium. Plan 1-2 pieces.
- `fanout_count 0-1`: Low. Combine with other entries or skip.

**Cross-entry fan-out**: Look for entries across repos that tell a connected story. Example:
- qortex "Mnemosyne Phase 1" + buildlog "Rule-Level Attribution" → Series: "Teaching Your AI to Learn"

### Step 4: Identify Content Opportunities

For each repo's recent activity + buildlog entries, identify narratives in these categories:

| Category | What to Look For | Audience |
|----------|-----------------|----------|
| **Provocation** | Gaps you found in the industry. Claims nobody else is making. | LinkedIn + blog |
| **Architecture** | Novel technical decisions. Why X not Y. The tradeoff. | Blog (technical) |
| **Build Log** | Velocity stories. "Here's what I shipped and how." | Substack + blog |
| **Tutorial** | Something you built that others could learn from. | Blog + HN |
| **Results** | Data. Measurements. Before/after. Convergence curves. | Blog + paper |
| **Personal** | "I noticed X while using my own tool." The visceral moment. | LinkedIn |

For each opportunity, extract:
- **Title**: Punchy, specific, non-generic. "Every AI Framework Has Memory. None of Them Learn." not "My Thoughts on AI Memory."
- **Source material**: Which PRs, issues, code files, buildlog entries, and commits provide the raw material
- **Links**: Full GitHub URLs to PRs, issues, and relevant code paths
- **Buildlog citation**: If sourced from a buildlog entry, include `buildlog://{repo}/{slug}` reference
- **Series**: Does this belong to a multi-part arc? Which one?
- **Channel**: LinkedIn (300-400 words), Blog (1000-3000 words), Substack (800-2000 words), or HN submission
- **Dependencies**: Does this need data/results that don't exist yet? Or can it be written now?

### Step 5: Check Existing Plan

Before generating output, check if a content plan already exists:

```bash
ls "${VAULT}/Writing/"
```

From Peleke/brunnr, MIT.

More from Peleke/brunnr