Search And Scale Os

Search & Scale OS

The orchestrator. It doesn't do the work itself. It reads where the search stands, tells the user the one highest-leverage thing to do next, routes to the skill that does it, and holds them to a weekly cadence. Think of it as the operating system every other skill plugs into.

When to use

• Starting out: "help me run my job search," "set this up," "search-and-scale-os."
• Any check-in: "where am I at?", "what should I do next?", "what's the highest-leverage thing right now?"
• The weekly ritual: "run my weekly review," "how did this week go?"
• Whenever the user is unsure which skill to invoke. This routes them.

Inputs

• Reads: workspace/profile/ (is setup done?), workspace/pipeline/tracker.csv (full pipeline state via application-tracker), the application records under workspace/pipeline/applications/, and workspace/output/ (recent artifacts).
• Needs from user: nothing to check status. For a weekly review, optionally their available hours this week and any constraints (deadlines, target start date) to calibrate targets.

Workflow

1. Setup check: is the foundation there?

If workspace/profile/ is empty or missing (no target-roles.md / target-companies.md / master resume) → stop and route to job-search-setup first. Everything downstream scores and tailors against the profile. Without it, the OS is guessing. Don't proceed to status until setup exists. (A missing stories/story-bank.md is not a setup blocker. Stories are built by the separate story-bank skill. Surface it as a recommended next action, not a bounce back to setup.)

2. Read pipeline state (via application-tracker)

Load tracker.csv and surface, concisely:

• Funnel snapshot. Counts by status: found · researching · tailoring · applied · screening · interviewing · final · offer · negotiating · (and terminal: accepted/rejected/withdrawn/ghosted).
• Priority-A next actions. Every A-priority row whose next_action_date is today or overdue, most urgent first.
• Stale / overdue. Rows with a next_action_date in the past, or ones stuck in one status too long (e.g. applied over 14 days with no movement means follow up or mark ghosted, and interviewing with no scheduled next_action means chase).
• Top-of-funnel health. How many active A-fit roles are in play vs. applied. Thin top-of-funnel is the most common silent failure.

3. Recommend the single next best action

One recommendation, not a menu. Diagnose where the search is constrained and route to the skill that unblocks it:

Lead with "Here's what I'd do next:" then the one action, why it's highest-leverage now, and the skill to run. Name any A-priority overdue items that must be cleared regardless.

4. Weekly review cadence

When asked (or when a week has passed since the last one), run the review:

A. Activity vs. targets. Did the week hit baseline volume? The default weekly targets (scale to the user's hours, and treat these as the floor for a serious full-time search):

• ≥ 5 new A-fit roles sourced (via job-sourcing → job-fit-scoring).
• ≥ 3–5 applications submitted, each with a referral path attempted first (networking-and-referrals before resume-tailor/cover-letter).
• ≥ 10 outreach touches (new intros, follow-ups, warm-network nudges).
• All scheduled interviews prepped (company-research + interview-prep) and all follow-ups sent within 24h (interview-followup).
• ≥ 2 LinkedIn touches (post/comment/engage) if running linkedin-presence.

Show actual vs. target per line, green/yellow/red.

B. Funnel conversion check. Compute stage-to-stage rates and flag the weak joint:

• applied → screening (low means résumé, targeting, or no referrals. Fix resume-tailor, job-fit-scoring, networking-and-referrals)
• screening → interviewing (low means screen prep or positioning. Run interview-prep)
• interviewing → final/offer (low means interview execution. Run interview-prep, story-bank)
• A high found→applied drop (roles scored but never applied) means execution stall, not a pipeline problem. Diagnose the single worst-converting joint and tie next week's focus to it.

C. Retro, what worked and what to change. 2–3 lines each:

• What produced real movement (a referral that landed a screen, a message that got replies) → do more.
• What ate time with no return (cold aggregator applies, over-tailoring low-fit roles) → cut.
• One process change for next week.

D. Set next week's plan. Restate the targets, name the specific A-priority actions, and update next_action_dates in the tracker.

5. Coach the rhythm

Reinforce the two principles that separate a search that lands from one that stalls:

• Consistent weekly volume. A steady 5 per week beats a 30-in-one-weekend binge followed by three dead weeks. Pipelines are lagging indicators. Today's volume is next month's interviews.
• Referral-first applying. A referred application converts at multiples of a cold one. Every A-role gets a referral path attempted before it's submitted cold. Spray-and-pray is the default failure mode. The OS exists to prevent it.

Lifecycle → skill routing map (all 17)

The full system, in the order a search moves through it:

Output

• Writes to: updates workspace/pipeline/tracker.csv (next_action / next_action_date on the rows it plans) via application-tracker, and optionally appends a dated entry to workspace/output/weekly-review.md.
• Format: a status readout (funnel snapshot + A-priority/overdue list), the single next-best-action recommendation with the skill to run, and, on review weeks, the activity-against-target table, conversion check, retro, and next-week plan.

Rules & quality bar

• Give a single next action and commit to it. Diagnose the bottleneck and name the single highest-leverage next move. Lead with "Here's what I'd do next."
• Setup gates everything. No profile means routing to job-search-setup. Don't run the rest on guesses.
• Never applies, sends, or submits. The OS plans and routes. A human approves every outbound artifact. No auto-apply and no auto-send, enforced at every downstream skill.
• Referral-first is a rule, not a nicety. Flag any application heading out cold with no referral attempt.
• Hold the cadence honestly. Report actual against target even when the week missed. The retro is where a stalling search gets corrected.
• No fabrication anywhere downstream. The OS inherits and enforces the system-wide rule, and gaps stay visible.

Related skills

• Routes to all 16 others. See the map above. Reads pipeline through application-tracker. First stop for every session. Last stop is the weekly review.