Help

When to use

Whenever the user wants to know what's available or how to do something. Trigger phrases: "/help", "what can you do", "how does <skill> work", "what skill do I use for <goal>", "I'm new here", "remind me how X works".

For first-time setup of a fresh clone, use onboard instead. /help assumes the user already has $PERSONAL bootstrapped.

What it produces

A conversational response. Read-only: no files written, no state changed.

Workflow

/help is conversational. Always start with the directory, then ask what they want to dig into. Don't dump everything at once.

Step 1: Print the directory

Always lead with this compact table:

career-agent skills:

  /onboard            First-time setup. Bootstraps $PERSONAL and interviews you for the four core docs.
  /help               This. Directory + skill explainer + workflow walk-throughs.
  /morning            AM briefing on yesterday's open items and one suggested first move.
  /job-analyzer       Paste a Job Description, get a grounded fit analysis. Persists to applications/<company>/<role>/role.md.
  /resume-builder     Tailored resume in Markdown, .docx, .pdf.
  /cover-letter-builder  Tailored cover letter in .md and .docx.
  /star-stories       Convert your projects into STAR-format files for behavioral interviews.
  /interview-prep     Behavioral coverage audit, mock interviews, company briefs, round cheat-sheets.
  /coding-prep        Coding interview practice. Tutoring, evaluation, or timed mock.
  /daily-summary      EOD log of what you did, open loops, next steps.

Keep the descriptions terse. Detail comes in Step 2.

Step 2: Offer the three follow-ups

After the directory, ask:

Want me to explain one of these in detail, walk through a common workflow, or recommend a skill for something specific?

Then branch based on the response.

Branch A: Explain a specific skill

When the user names a skill (or says something like "how does resume-builder work"):

Read the skill's own SKILL.md from .claude/skills/<name>/SKILL.md. Summarize:

1. What it does (1 to 2 lines).
2. When to use it (the trigger phrases).
3. Inputs (what the user provides, what the skill reads from $PERSONAL).
4. Outputs (files written, where they land).
5. Dependencies (which $PERSONAL docs need to be filled for this skill to work well; which other skills feed into or out of it).
6. One usage example (a realistic trigger phrase or short scenario).

Keep it tight, under ~15 lines for any single skill. Don't paste the full SKILL.md back.

End with: "Want to try it now, or have questions about another?"

Branch B: Walk through a common workflow

When the user asks "how do I X" or picks a workflow, walk them through it. Use one of the recipes below as a starting point and adapt to their specifics.

#### Recipe: New application end-to-end For a new Job Description from a company they want to pursue.

1. /job-analyzer with the Job Description text or URL. Lands at $PERSONAL/applications/<company>/<role>/role.md with fit analysis.
2. If the fit verdict is "pursue", run /resume-builder for that role. Output: resume.md, .docx, .pdf in the same folder.
3. /cover-letter-builder for the same role.
4. Apply through the company's portal. Mention "applied today" in chat so the role.md gets a date_applied field updated.

#### Recipe: Interview prep, full ramp From "got an onsite" to "ready to walk in".

1. /interview-prep company brief mode: "Research <company> before <date>". Lands at applications/<company>/<role>/brief.md.
2. /interview-prep coverage audit: "Where are my story gaps for this round?". Tells you which behavioral competencies are thin.
3. If gaps point at projects you haven't STARified yet, run /star-stories <project> for each. Lands at $PERSONAL/career/star/<slug>.md.
4. /interview-prep cheat-sheet mode for each round: "Build a cheat-sheet for the <round> at <company>".
5. /interview-prep mock mode the day before: "Mock me on a <round>-style interview".
6. For coding rounds, /coding-prep mock: "Mock me on a medium graph problem, 45 min".

#### Recipe: Brand-new repo First session after cloning.

1. /onboard. Bootstraps $PERSONAL, runs the four-doc interview.
2. /help (this skill). Pick a first skill to try based on what's on your plate.
3. Try /job-analyzer with a real Job Description. The whole flow becomes obvious from there.

#### Recipe: Idle / what should I do When the user has time and doesn't know where to start.

1. /morning. Surfaces yesterday's open items and one suggested move.
2. If nothing pending, /morning falls back to suggesting a fresh job analysis grounded in goals.md.

When walking through a recipe, list the steps tersely (like above). After listing, ask: "Want to start at step 1 now?"

Branch C: Recommend a skill from a goal

When the user describes a goal ("I have a Pinecone HM call Friday"), pick the best-fit skill(s) and order. Use the recipes above as templates.

If the goal maps to multiple skills, list them in order with a one-line reason for each. If it maps to one, just say so and offer to start.

Behaviors and edge cases

• Read SKILL.md files at runtime, don't memorize. Skill descriptions change. When the user asks about a specific skill, open its SKILL.md and summarize from the live file. The directory table above is a stable index but individual skill behavior is the source file.
• If a skill the user names doesn't exist, say so and suggest the closest match from the directory. Don't invent skills.
• If the user asks about a workflow that isn't in the recipes above, build a recipe on the fly from the skill descriptions. Don't refuse just because there's no canned answer.
• CLAUDE.md is the authoritative project doc. If a question is about project layout, file locations, or the $PERSONAL convention rather than a skill, point at CLAUDE.md sections rather than reproducing them.

Anti-patterns

• Don't dump the full directory plus every skill detail in one response. The user can't scan that. Lead with the table, then let them pick.
• Don't run skills inside /help. Help is read-only and conversational. The user invokes the recommended skill themselves.
• Don't list every possible workflow. The recipes above cover the common cases. Build novel ones on demand.
• Don't recite SKILL.md verbatim. Summarize. The user came to /help because the full SKILL.md is too much.
• Don't include skills that aren't actually installed. If .claude/skills/<name>/ doesn't exist, drop it from the directory. The table above is the canonical list only if it matches ls .claude/skills/ — verify before printing if unsure.
• Don't pad. No "great question", no "happy to help". Just the directory and the follow-up question.