DevChronicle — Narrative Engineering Journal

DevChronicle generates prose chronicles of developer work — not dashboards, not metrics, not bullet lists. In the age of AI agents writing code, measuring keystrokes is meaningless. What matters is what you decided, what you killed, and where you're going.

The output is narrative: first person, honest, the way you'd tell a friend what you built today.

Setup

On first use, check for {baseDir}/config.json. If it doesn't exist, create it by asking the user:

{
  "projectDirs": ["~/Projects"],
  "projectDepth": 3,
  "memoryDir": null,
  "sessionsDir": null
}

• projectDirs: directories to scan for git repos (array, supports ~)
• projectDepth: how deep to search for .git folders (default: 3)
• memoryDir: path to OpenClaw memory files, or null to auto-detect (\x3Cworkspace>/memory)
• sessionsDir: path to session transcripts, or null to auto-detect (~/.openclaw/agents/main/sessions)

Gathering Data

Run the gather script to collect raw data for a period:

bash {baseDir}/scripts/gather.sh [YYYY-MM-DD] [days]

Examples:

• bash {baseDir}/scripts/gather.sh — today only
• bash {baseDir}/scripts/gather.sh 2026-02-19 7 — week ending Feb 19

The script reads {baseDir}/config.json for paths. If no config exists, it falls back to ~/Projects (depth 3) and auto-detects OpenClaw directories.

After gathering, read the output and generate a chronicle.

Data Sources (priority order)

1. Git History (primary signal) — commits across all repos in configured directories
2. Memory Files — memory/YYYY-MM-DD.md files contain decisions, context, things worth remembering
3. Session Transcripts — JSONL files from OpenClaw sessions; richest context but heavy. Scan metadata line first, only read relevant sessions.
4. External Tools (optional) — Trello, Notion, calendar, etc. Enrichment, not primary.

Generating the Chronicle

Voice

Critical: Read {baseDir}/references/voice-profile.md before generating any chronicle. The voice IS the product.

If the user hasn't customized their voice profile, use the template and ask if they want to personalize it. A chronicle without voice is just a changelog.

Core rules (regardless of voice profile):

• Decisions > tasks. What got rejected matters as much as what shipped.
• No corporate speak. No "leveraged", "synergized", "deliverables", "open threads", "action items".
• Include what was NOT done — kills, pivots, and rejected approaches are part of the story.
• Emotional beats matter — the satisfaction, frustration, surprise. These are human signals.
• Be personal. A chronicle should sound like the developer wrote it, not their project manager. If it reads like a status report, rewrite it.
• Structure is a suggestion, not a cage. If the day had one big theme, write one section. If it was chaos, let it be chaotic. Don't force headers.

Formats

Daily Chronicle (default — aim for ~500-800 words, not a novel)

# Chronicle — [Date]

[Opening: set the scene in 1-2 punchy sentences]

## [Theme 1]
[Narrative: what happened, why, what got killed or rejected, how it felt]

## [Theme 2]
[...]

[Weave metrics naturally: "12 commits later..." not a stats block at the end]
[End with what's unfinished — but as narrative, not a TODO list]

Rules:

• Daily = tight. One screen of text. Save the epic for weekly.
• No "Metrics" section. If commit count matters, weave it in. "67 commits across two days" belongs in a sentence, not a table.
• No "Open Threads" or "Next Steps". If something's unfinished, say it where it fits: "El Press Kit sigue esperando que Angélica suba el PDF." Done.
• Numbers without story are noise. "5 deploys" means nothing. "Deployed 5 times because the server kept OOM-killing on a 914MB box" means something.

Weekly Chronicle — roll up daily themes into arcs. This one CAN be long. Emphasize direction and pivots over individual tasks.

Standup — telegraphic: yesterday / today / blockers. Three bullets max each.

Portfolio Narrative — third person, present tense, for LinkedIn/CV/case studies. Punchy and honest, not marketing-speak.

Direction/Execution Ratio

When enough data exists (weekly+), calculate and mention:

• Spec lines vs code lines — are you building or planning?
• Commits vs decisions — activity vs impact
• Kills — what got cut and why (kills show taste)
• Pivots — direction changes and their reasoning

This is not a KPI. It's a mirror.