Agent skill
reengagement-segmenter
Segment a lapsed audience by the actual blocker that stalled each relationship, then plan a distinct reengagement motion per segment, including the segment that should be left alone.
Filed under Outbound email.
From esmefong/gtm-skills · 8 skills · 0 · pushed 2026-09-04
What it does when it runs
Segment a lapsed audience by the actual blocker that stalled each relationship, then plan a distinct reengagement motion per segment, including the segment that should be left alone. Works on dormant pipeline, lapsed customers, cold trial users, or an inactive list. Applies a blocker taxonomy (never activated, timing, lost to incumbent, price, trust damage, drifted, unknown), enforces hard exclusions first (unsubscribed, complaints, active competitor contracts), flags records whose stated reason contradicts their evidence, and drafts sequences through the outbound-engine gate. Use whenever the user says "win-back", "reengagement", "re-engage", "lapsed customers", "dormant accounts", "stalled deals", "gone quiet", "cold pipeline", "churned users", or wants to revive a list without blasting it. Consumes icp-research and brand-voice-guide; sequences inherit outbound-engine's no-send discipline.
Read from the skill and the 9 files bundled beside it. A skill’s own description is written to be selected by an agent, so it describes the job and not the dependencies.
- Keys and connectors you must supply
- None found.
- Hosts it reaches
- www.techtimes.com
- Tool permissions it declares
- No
allowed-toolsin the frontmatter. It does act, so it runs under whatever permissions your session already grants. - Actions present in the files
- writes files
Install it
View source on GitHub ↗git clone --depth 1 --filter=blob:none --sparse https://github.com/esmefong/gtm-skills.git /tmp/gtm-skills git -C /tmp/gtm-skills sparse-checkout set "reengagement-segmenter" mkdir -p ~/.claude/skills/reengagement-segmenter cp -R "/tmp/gtm-skills/reengagement-segmenter/." ~/.claude/skills/reengagement-segmenter/
Picked up without a restart. A project skill of the same name is shadowed by your personal one. For one repository only, swap ~/.claude/skills for .claude/skills. Claude Code docs ↗
The folder is the same in every client that implements the format — 46 of them — so if yours is not above, only the destination changes.
The skill
Source on GitHub ↗Reproduced in full from esmefong/gtm-skills/blob/99400e67cdf7535d05377d7ca279ff81d550525e/reengagement-segmenter/SKILL.md, which is licensed MIT (repository). 832 words, 11 headings.
Reengagement Segmenter
Turn "the lapsed list" into distinct audiences with distinct reasons for silence, and treat the reason as the segment. The design rests on one argument: win-back blasts fail because they treat silence as one condition, when silence is the least informative signal a contact can send. A deal stalled at security review, a user who never reached value, and a buyer who signed with the incumbent all look identical in a recency report and need three different messages, or none.
The other half of the argument is the repo's standing one: the most valuable output of a reengagement pass is often the list of people you decide not to contact, documented.
What this consumes
| Input | From | What it provides |
|---|---|---|
| The lapsed records | CRM export, deal notes, product events, support history | The evidence each blocker call rests on |
| ICP | icp-research | Disqualifiers (bad-fit lapse is leave-alone, not win-back), triggers for the why-now |
| Voice directory | brand-voice-guide | The register sequences are written in |
| What changed on your side | The user | Reengagement needs a why-now; "we miss you" is not one |
The last row is the load-bearing one. A reengagement sequence exists because something changed: your product closed the gap that stalled them, their situation visibly shifted, or a trigger from the ICP fired. No change, no sequence; that cohort waits.
The workflow
Five phases. Show the user the segment map (end of Phase 3) before any drafting.
Phase 1: Assemble the evidence, not just the list
Pull the lapsed records with whatever evidence exists per record: last meaningful activity, lifecycle stage reached, stated reasons (closed-lost fields, cancellation surveys), and observed behavior (support tickets, usage before silence, emails). First-party evidence is messy and contradictory; that is expected and handled in Phase 3, not smoothed over.
Phase 2: Run the exclusions first
Before any blocker reasoning, apply the hard gates from references/blocker-taxonomy.md:
unsubscribed or do-not-contact, complaint history, active competitor contracts, ICP disqualifiers,
and the dormancy horizon. These records go to the leave-alone lane with reasons, which ships
in the deliverable. Reengaging someone who opted out is not a growth motion, it is a compliance
incident.
Phase 3: Assign blockers, honestly
Tag each surviving record with a blocker from the taxonomy, citing the evidence. Three rules:
- The evidence outranks the label. When the CRM says "price" but the thread shows a security
stall, the record is flagged as a conflict and resolved by judgment, with the resolution
written down.
scripts/segment_audience.pydetects these conflicts mechanically. - Unknown stays unknown. A record with no blocker evidence goes to the diagnostic segment, whose sequence asks one honest question instead of assuming a reason.
- The script does the bookkeeping. Deterministic rules (exclusions, recency bands, tag precedence, conflict detection) run mechanically; the judgment calls are yours and get recorded.
Phase 4: Plan one motion per segment
Each blocker has its own sequence archetype in the taxonomy: the never-activated get the value they missed, the timing-stalled get a low-pressure trigger watch, the security-stalled get the specific thing that changed, the diagnostic segment gets one question. Two universal rules: reengagement sequences are shorter than cold ones (two to three touches), and every sequence opens with the why-now change, not with an apology for the silence.
Phase 5: Draft, gate, package
Draft in the brand voice and run every sequence through outbound-engine's
sequence_gate.py: the freshness check applies to the why-now change, the slop list catches
"just checking in" (the canonical reengagement failure), and the voice Avoid list holds. Package
per references/output-template.md: the segment map with counts, the per-segment plans, the
conflict log, and the leave-alone lane with reasons.
Scripts
| Script | What it does | Network | Keys |
|---|---|---|---|
scripts/segment_audience.py | Deterministic segmentation: exclusions, recency bands, blocker tags, conflict flags, counts | None | None |
Failure modes to avoid (the quality bar)
- One blast to the whole list. The disease this skill treats.
- "Just checking in." The outbound gate catches it mechanically; do not write it.
- Guessing the blocker. No evidence means the diagnostic segment, not your best story.
- Reengaging the excluded. Unsubscribes and complaints are permanent exits from this motion.
- Apologizing for the silence. Open with what changed, not with the gap.
- A hidden leave-alone lane. The records you chose not to contact, with reasons, are output.
- Trusting the CRM label over the thread. Stated reasons are hypotheses; conflicts get logged.
Bundled references
references/blocker-taxonomy.md: the segments, the evidence each requires, the exclusion gates, and the sequence archetype per segment. Read at Phases 2-4.references/output-template.md: the audience.json schema the script consumes and the segment report format. Read at Phases 3-5.examples/cohere/: a dormant-pipeline pass in Cohere's voice on synthetic records (fictional counterparties, labeled, with real contradictions), the segment map, and a gated sequence for the security-stall segment.
Files bundled with it
These load only when the skill asks for them, so they cost nothing until it runs.
Need help setting it up?
This page tells you what reengagement-segmenter does and what it needs. Cheetah builds the agent setup it runs inside: data, CRM, sequencing and the guardrails.
Book a call →The directory stays free. There is nothing gated behind this.