── SKILL
refresh-orchestrator✓ APPROVED

Stop rebuilding data that did not move.

The cheap gate that decides, before an expensive workflow runs, whether to skip, partially refresh, recycle, or fully regenerate.

Inside: The four verdicts, the section taxonomy, and a dollarized savings estimate.

01 / HOW TO USE

Install it in one line, or paste it in.

1
In Claude Code (one command)
Copy the install line, paste it into your terminal, and restart Claude Code. The skill installs itself to ~/.claude/skills/ and runs automatically when it is relevant.
2
In Claude, ChatGPT, or a Project (no terminal)
Open the file, then upload it to your chat or paste its contents in. A skill is just a markdown file of instructions, so any capable AI can follow it.
New to skills? A skill is a plain-text file that teaches your AI a workflow. Point any capable assistant at it and it follows the steps, on your data.
02 / MAKE IT YOURS

Connect your context. Set it to your motion.

CONNECT YOUR CONTEXT · AND WHY IT HELPS
your data sources

it reads last-modified timestamps and field-level deltas automatically, so the verdict is evidence-backed, not eyeballed.

a meeting or activity tool

catches new events since the last run that should force a refresh.

a product-analytics tool

catches a usage-trend shift that should re-open a section.

SET IT TO YOUR MOTION

This was built to gate a suite of GTM artifacts. Set these to your stack:

Set thisWhat it isDefault / Example
ARTIFACT TYPESwhat you regenerateaccount briefweekly reportdeal reviewpipeline analysis
DATA SOURCESwhere "did it change" comes fromCRM last-modifiedmeeting countusage trend
STATIC SECTIONSreuse, stable 30+ dayscompany researchbuying committeetech stack
DYNAMIC SECTIONSrebuild every runactivitycurrent dealstime-bound cards
CONDITIONAL SECTIONSrebuild only when a trigger fireschampion changefit-score deltausage-tier shift
SKIP_DAYSage under which unchanged means skip7
FULL_DAYSage over which force a full rebuild30
03 / THE FULL SKILL

Everything the skill does, in full.

── WHAT THIS DOES

Before you spend tokens rebuilding a brief, report, or analysis, this checks whether anything underneath it actually changed. It returns one of four verdicts (skip, partial, recycle, or full), a section-by-section plan of what to reuse and what to rebuild, and an estimate of what you save by not regenerating the whole thing. It is meant to be cheap, so it can gate expensive work and pay for itself many times over.

── THE METHOD
  1. 1
    The four verdicts

    Every artifact gets exactly one:

    • SKIP: age at or under SKIP_DAYS and no underlying data changed. Return the existing artifact, rebuild nothing, cost zero.
    • PARTIAL: recent, but some sections have stale underlying data. Rebuild only those sections and splice them into the existing artifact. Typical savings 60 to 80 percent.
    • RECYCLE-CORE: older, but the identity and static sections are unchanged while the dynamic ones are all stale. Reuse the static cards, rebuild the activity-driven ones. Typical savings 30 to 50 percent.
    • FULL: older than FULL_DAYS, or an identity-level change (acquisition, segment shift, recategorization), or no existing artifact. Run the whole workflow. Cost 100 percent baseline.
  2. 2
    Static vs dynamic sections

    Every artifact type splits three ways. STATIC sections are identity-grade and stable for 30+ days (company research, buying committee, tech stack, history). DYNAMIC sections rebuild every run (days-dark, current deal health, time-bound wins, coverage math). CONDITIONAL sections rebuild only when a named trigger fires (a champion job-change, a fit-score delta past threshold, a usage-tier shift). Knowing which section is which is the whole skill; map yours once and the plan writes itself.

  3. 3
    Freshness signals

    The verdict reads: artifact age; last-modified on the source record; field-level diffs mapped to the sections each field drives; new activity since the last run; a usage-trend tier shift; a fit-score delta past a threshold. Each signal points at specific sections, not the whole artifact, which is what makes a partial plan possible.

  4. 4
    The savings estimate

    Every partial or recycle plan states the estimated token reduction and cost saved versus a full rebuild, so the call is auditable and anyone can see when the gate is being too aggressive.

── QUALITY GATES
  • Evidence-first verdicts. Never "the brief is stale." Always "stale because the source changed on {date} and the artifact is from {date}."
  • No false-fresh calls. If the evidence is ambiguous (you cannot read the artifact or the source), default to FULL. Better to spend than ship stale.
  • Conservative on identity changes. Any identity-level signal (acquisition, segment shift, recategorization) forces FULL regardless of age.
  • First run of any new artifact is always FULL. There is nothing to reuse yet.
── OUTPUT (EXAMPLE)
EXAMPLE OUTPUT
REFRESH PLAN - account brief for acme.com
Verdict: PARTIAL   (artifact 4 days old)
Reuse:      company research, buying committee, tech stack, history
Rebuild:    days-dark, deal health, top wins
Evidence:   source last-modified 2 days ago; 2 new meetings; usage trend flat
Savings:    ~73% fewer tokens vs a full rebuild
── WHERE THE NUMBERS COME FROM

SKIP_DAYS (7), FULL_DAYS (30), and the fit-score delta threshold are defaults from one team's cadence, not laws. A weekly artifact and a quarterly one want different windows. If your work changes slowly, widen the windows; if it moves daily, tighten them. The four-verdict logic does not change. The thresholds are yours.

── EXAMPLE CHAT
EXAMPLE CHAT
You
Do we need to fully regenerate the acme.com account brief, or can we skip it? Last one ran 4 days ago.
CLAUDE
Checked it against the source. Partial rebuild, not full. REFRESH PLAN - account brief for acme.com Verdict: PARTIAL (artifact 4 days old) Reuse: company research, buying committee, tech stack, history Rebuild: days-dark, deal health, top wins Evidence: source last-modified 2 days ago; 2 new meetings; usage trend flat Savings: ~73% fewer tokens vs a full rebuild The catch worth knowing: I don't call this fresh just because it's recent. I call it PARTIAL because I can point to exactly what changed, 2 new meetings, source modified 2 days ago, and what didn't. If I couldn't see that evidence, the default flips to FULL, because shipping stale work costs more than the tokens.
You
What would push this to a full rebuild instead?
CLAUDE
Any identity-level signal, an acquisition, a segment shift, a recategorization, forces FULL no matter how fresh the artifact looks. Age alone doesn't do it, the account changing who it is does. Sharpener: connect your CRM's last-modified timestamps directly and I'll prove every verdict against real dates instead of the note you pasted.
04 / GO FURTHER

Where an operator takes this next.

The manual check proves the gate is worth running. Here is the version that runs itself.

The gate pays for itself the first week it stops a rebuild nobody needed.

01
Gate every report behind this first

Wire it in front of your weekly revenue report or account-brief generator so nothing regenerates until this verdict says so.

02
Watch for identity triggers automatically

Connect Salesforce so an acquisition or segment-shift field change forces FULL the moment it's logged, not at the next scheduled run.

03
Track the savings over a quarter

Log every verdict and its token estimate to a sheet so you can show finance what the gate is actually worth.

05 / PART OF A BIGGER PLAY

One skill is the on-ramp.

── GET NEW SKILLS IN YOUR INBOX

One operator-built skill a week. Free.