suggest-skills

Suggest Skills

Compares ground-truth to demand signals from the target (role / company / domain) and surfaces gaps with priority.

Inputs

$ARGUMENTS:

• Optional: --target-role=<title> — comma-separated titles.
• Optional: --target-company=<slug>.
• Optional: --target-domain=<slug>.
• Optional: --horizon=<weeks> (default 12) — time the user has to close gaps.

At least one target required.

Examples

/career:suggest skills --target-role="staff engineer, ai platform"
/career:suggest skills --target-domain=ai-evals --horizon=8
/career:suggest skills --target-company=snowglobe

Procedure

1. Load context

• ground-truth.md (current skills + freelance-brand if relevant).
• Target source(s): postings for --target-role, briefs for --target-company, note for --target-domain.

2. Extract demand signals

For each target source, list the skills mentioned with frequency. Group:

• Hard skills (languages, frameworks, tools).
• Domain skills (e.g., "RLHF", "compliance frameworks").
• Soft / scope skills (e.g., "leading 0→1", "vendor management").

3. Compare against ground-truth

Classify each demand-signal skill:

• have — clearly in ground-truth.
• partial — adjacent or unproven.
• gap — absent.

For each gap and partial, estimate:

• Closing effort (low / med / high) within --horizon.
• Priority (must / should / nice) based on how often it appears across target sources.

4. Write the recommendations file

# Skill gap analysis — {{date}}

> Targets: {{role|company|domain summary}}
> Horizon: {{weeks}}
> Ground-truth as of: {{ground-truth-mtime}}

## Must close (within horizon)

| skill | priority | effort | what evidence would count | suggested approach |

## Should close (over horizon)

| skill | ... |

## Nice to have

| skill | ... |

## Already strong (don't oversell, don't ignore)

| skill | where to demonstrate it |

## Sources

1. ...

5. Print summary

recommendations: recommendations/skills-<date>.md
must: <n>
should: <n>
nice: <n>

next: /career:suggest projects --close-gap=<skill>

Guardrails

• "Evidence that would count" must be concrete (a project, a write-up, a contribution) — not "learn X".
• Don't recommend closing every gap. The horizon is real; surface the trade-off.
• partial is honest. Don't classify as have to flatter the user; don't classify as gap to manufacture work.

Failure modes

• No target source available → bail; ask user to specify target.
• ground-truth too thin → run /career:ground-truth edit first.