Daily Summary

Daily Summary

When to use

At end of day, when the user wants a record of what happened in this tool and what's still open. Trigger phrases: "wrap up the day", "summarize today", "EOD", "end of day".

For morning briefings on yesterday's carryover, use morning instead.

What it produces

A markdown file at $PERSONAL/career/daily-log/YYYY-MM-DD.md (today's date, ISO format) with the sections below. $PERSONAL is the personal-docs sibling directory defined in CLAUDE.md; resolve it once at the start of the session. Omit any section that has nothing real to report — don't fill with placeholder text or generic advice.

Section 1: What I did

Artifact-level log of today's activity. Detect by:

• Scanning $PERSONAL/applications/ for files with mtime = today
• Reading the conversation transcript for skill invocations and concrete decisions
• Checking for new/modified entries in $PERSONAL/career/coding-log.md, $PERSONAL/career/story-bank.json, etc.

Examples:

• "Analyzed Stripe Staff Payments Job Description → $PERSONAL/applications/stripe/staff-payments/role.md"
• "Generated tailored resume for Anthropic MTS role → $PERSONAL/applications/anthropic/mts/resume.pdf"
• "Mock interview (Google E5 style, 45 min) — verdict: lean hire, weak on graphs"
• "Practiced 3 sliding-window problems — struggled on minimum-window-substring"

Each bullet should reference a concrete file path or specific outcome, not a vague "we discussed X."

Section 2: Open loops

Work started today (or earlier) that isn't finished. Detect by (all paths under $PERSONAL):

• role.md exists with no resume.md next to it → tailored resume not yet built
• Mock interview happened but no follow-up practice on noted weak areas
• A role analyzed but no decision recorded ("save", "skip", "stretch")
• TODO-style notes in role.md files

Cite the specific file each open loop refers to.

Section 3: Patterns from today

Include this section only if ≥2 similar activities happened today. Look for cross-cutting signal:

• Recurring required skills across multiple Job Descriptions ("3 of 3 Job Descriptions flagged distributed systems depth")
• Recurring weakness across problems ("hesitated on 2 graph traversal problems")
• Comp range observations across postings

If only 1 activity happened, skip this section entirely. Forced patterns from a sample of 1 are noise.

Section 4: Next steps

Concrete actions, each tied to a specific artifact or upcoming event. NOT vague career advice.

Good examples:

• "Build tailored resume for Stripe Staff ($PERSONAL/applications/stripe/staff-payments/)"
• "Practice 2 medium graph problems before Friday's Anthropic onsite"
• "Add 'conflict with stakeholder' STAR story — gap revealed in today's mock"

Bad examples (don't write these):

• "Keep applying"
• "Stay focused on practice"
• "Review your resume"

Workflow

1. Auto-generate the summary from file mtimes and the conversation transcript. Don't ask the user any questions up front.
2. Write the file to $PERSONAL/career/daily-log/YYYY-MM-DD.md. If the file already exists (user invoked the skill twice today), append a ## Updated at HH:MM block instead of overwriting — preserves the earlier record.
3. Print the full summary to the terminal so the user can review.
4. Ask once at the end: "Anything I missed — decisions, recruiter calls, applications submitted outside Claude, or anything else worth recording?" Append their freeform response under ## Additional notes. If they say no, skip and finish.

File format

---
date: 2026-05-11
artifacts_created: 2
artifacts_modified: 1
---

## What I did
- ...

## Open loops
- ...

## Patterns from today
(only if ≥2 similar activities)

## Next steps
- ...

## Additional notes
(only if user added context after the prompt)

Anti-patterns

• Don't motivate or cheerlead. No "great progress today", no productivity scores. CLAUDE.md is clear: no padding.
• Don't turn this into an application tracker. Pipeline state (stages, dates applied, outcomes) is tracked optionally on each role.md. Daily summaries reflect what happened today, not pipeline state.
• Don't fake precision. No "you spent 2 hours on coding" — that can't be measured reliably. No "you're 30% through your search" — meaningless.
• Don't include sections with no real content. If there are no open loops, omit the section. Empty sections are noise.
• Don't dredge up old context. Only today's activity + actionable open loops. Stale items belong in morning, not here.