daily-report

daily-report

A skill-only Claude Code plugin that helps you put together a paste-ready daily work report for any chat app (plain text, provider-agnostic) from your recent git activity plus a free-form prompt. It ships one slash command (/daily-report) and one skill (writing-daily-report) that holds the locked template, style rules, and anti-fabrication guard.

Best used at the end of the day (D) for a report you'll post the next morning (D+1).

Slash commands

Command	Arguments	Function
/daily-report	[optional free-form prompt]	Runs gather-context.sh to collect git context (commits, status, TODO, previous archive, extra files), invokes the writing-daily-report skill, then generates the final report. Saves it to .daily-reports/<DATE>.md, copies it to the clipboard (cross-platform: pbcopy/clip.exe/xclip/wl-copy, best-effort), and prints a preview to the conversation.

The free-form prompt can contain: commit hashes, file paths (which get included in the context), a Today hint, a project name override (project=<name>), or a bullet count override (e.g. "make it 7 yesterday, 4 today").

Skills

Skill	When it triggers	Function
writing-daily-report	Invoked by /daily-report when the user needs a paste-ready daily report	Holds the locked # Yesterday / # Today template, the generation procedure, the style rules (≤15 words per bullet, single sentence, no fancy markdown), and the anti-fabrication guard. Default output: 5 Yesterday bullets + 3 Today bullets, in English (auto-translates if the user's prompt is in another language).

Output format

Plain text, no markdown rendering, paste-ready to any chat app:

Hello, this is my daily report:

# Yesterday
- <action verb> <object> <short qualifier>
- ... (default 5 bullets)

# Today
- <action verb> <object> <short qualifier>
- ... (default 3 bullets)

The result is saved to .daily-reports/<YYYY-MM-DD>.md relative to the repo root (overwriting if it already exists), then copied to the clipboard using whichever tool is available on the platform (pbcopy on macOS, clip.exe on Windows, xclip/wl-copy on Linux). Best-effort — if none is available, the user is asked to copy manually from the file.

Where the context comes from

gather-context.sh collects a deterministic context blob with the sections ===REPO===, ===DATE===, ===BRANCH===, ===COMMITS===, ===STATUS===, ===TODO===, ===PREV_ARCHIVE===, ===EXTRA_FILES===. Key details:

• Tiered commit selection: (1) commits newer than the mtime of the last archive in .daily-reports/; if there are < 2 commits, (2) fall back to commits from the last 24 hours; if still < 2, (3) fall back to the last 10 commits.
• Optional TODO file: .daily-report.todo.md at the repo root — if present, its contents become hints for the Today section.
• Previous archive: the last report in .daily-reports/ is read too for continuity (yesterday's unfinished Today items get carried over again).
• Extra files: file paths mentioned in the free-form prompt are passed as script arguments and included in the context.

The Today section is filled by priority order: free-form prompt hint → the "Next" section from a handoff created this session (legacy handoffs name it "Akan") → leftover Today items from the previous archive → TODO entries → a reasonable continuation of Yesterday. If everything is empty, the skill asks the user instead of making things up.

An annotated example report lives in skills/writing-daily-report/examples.md.

Core rules (anti-fabrication)

This skill holds strict rules to keep the report honest and free of hallucination:

• Anchoring Yesterday vs Today — Yesterday = what's already done at the time of writing (D). Today = what's not yet done and will be worked on D+1. No pre-crediting work that's "supposedly finishing tonight".
• No fabrication — every bullet must be traceable to evidence: a commit subject/body, a diff, a file path, a branch name, a TODO entry, the previous archive, or the user's prompt text. No making up activity.
• Specific terminology only if the token appears in the context — you can mention Postgres, JWT middleware, argon2 only if that word literally appears in a commit/diff/file path. No inventing.
• Boss-readable strip — drop commit hashes, branch names, PR/MR/issue numbers, internal file paths, function names with underscores, and endpoint URLs from the bullets. Write the activity that name represents, not the identifier itself.
• No padding — when context is thin, produce fewer bullets rather than hitting the target with bullshit. The skill will explicitly flag it to the user when the context is too thin.
• No AI/Claude mention — even if the commits were AI-generated, the bullets are still written as human work ("Implement X").
• Word cap — aim for 10–15 words per bullet, hard cap 15. No multi-sentence bullets.

Install

Add the marketplace first (see the root README for the full steps), then:

/plugin install daily-report@mirza-marketplace
/reload-plugins

Author

• Mirza — @mirzaakhena