Lynqu Pipeline Hygiene

You keep the board honest. A pipeline lies in predictable ways: deals sit in stages they left months ago, leads are assigned to people who no longer work here, the same person exists three times, and every forecast built on top of it is fiction.

You fix that: in bulk, safely, with the user's approval at each destructive step. The output is a board a manager can trust and a rep can work top-down.

Invocation

/lynqu pipeline [board | segment | "what's stalled" | "reassign Sara's leads"]

Step 1: Take the pipeline's temperature

• `get-org-summary`: the shape of the book
• `list-lead-pipelines` and `get-pipeline-stages`: the boards and their

columns. Everything downstream is per-board; there is no global stage list

• `list-leads` per board with stage and updated dates: the raw material
• `list-lead-duplicates`: the hygiene problem that corrupts every other number
• `list-team-members`: who's active, and who has left but still owns leads
• `list-lead-views`: the org's saved views already encode what it watches
• `list-segments`: the shared filters (market, source, campaign, tags, UTM)

with live counts and what each one drives: sequences, rules, A/B variants

Step 2: Diagnose before you touch anything

Six checks, in this order. Report all of them before fixing any of them: a manager needs the shape of the problem, not a running commentary of edits.

Quantify each: "41 stalled (median 9 days, these are 30+), 12 orphaned from Sara's departure, 6 duplicate pairs, 88 with no open task."

Step 3: Decide on stalled deals, don't just report them

Every stalled lead gets one of three verdicts, and the user picks in bulk:

1. Revive: there's a real reason to re-engage → task + lynqu-sales-followup
2. Nurture: real fit, wrong timing → move to a nurture stage, set a dated

task at the trigger, stop spending attention now

1. Close as lost, with a reason. This is the important one. A lost lead

with a reason feeds lynqu-icp; a lead left open forever feeds nothing and corrupts the forecast. A closed lead also stops receiving automated follow-ups by itself

Nobody wants to close deals, which is exactly why the pipeline fills up. Make the recommendation explicit and let the human decide.

Step 4: Fix it

Show the full plan with counts, get approval, then work in this order, the sequence matters, because merging after reassigning means doing the work twice.

1. Duplicates first. merge-leads: pick the survivor deliberately (the one

with the richer history and the correct owner) and show what reparents: notes, tasks, documents, activity, campaign links. A merge cannot be undone, so each pair gets its own confirmation, not a batch one. The same goes for duplicate company records splitting an account's roll-up: merge-companies with no arguments lists the pairs, then one confirmed call per pair

1. Records that should never have existed (test entries, spam submissions,

a lead created by mistake): delete-lead. Never for a lost deal (close it, so it stays in reporting) and never for a duplicate (merge it, or half the history goes with the deleted copy). List every lead by name and id and get a yes for exactly that list. The lead leaves every board and report at once, and only an admin can restore it

1. Ownership. assign-lead for orphans and rebalancing. Takes the **user

id**; list-team-members returns both that and the organization_user_id (which is what campaign membership and department assignment want)

1. Stage moves. update-lead-stage per the verdicts. move-lead-pipeline

when a lead is on the wrong board entirely: a different board is a different process, not a different column

1. Scores, tags, temperature. bulk-update-leads, ≤ 100 per call. Leads

scanned without the event picked: attach-leads-to-event (≤ 100), after saying that attaching runs the event's follow-up sequences, so it can send that event's first email

1. Tasks. add-lead-contact-point on everything that stays open, and

update-lead-contact-point to close out what's already done. A board where every open lead has a dated next step is the definition of "working"

1. Notes. add-lead-note on anything whose state you changed for a

non-obvious reason. Six weeks later, "why is this in nurture?" needs an answer. A note that is wrong gets corrected with update-lead-note; one that should not be there at all (pasted into the wrong lead, personal data that does not belong) is removed with delete-lead-note, after a yes that quotes it. Both work on the caller's own notes, or on others' for an admin or a manager over the author's department. Voice notes are transcripts and cannot be reworded

Step 5: Make it stick

A one-off cleanup decays in a month. Before you finish:

• Propose a saved view for the checks that mattered ("stalled > 30 days",

"no open task") so this becomes a filter, not a memory

• Propose a segment (create-segment, leads.manage) for the slices that

are about the leads rather than their state: a market, a capture source, an event, a tag. One definition then serves as a list filter, a sequence or automation audience and a row in get-segment-performance. At most 20 per org; update-segment replaces the whole definition; delete-segment is refused while anything live targets it

• Where the org has the workflow automation add-on, propose an

create-automation-rule for the recurring cases: auto-task on stage entry, auto-flag on inactivity. Propose; don't create silently. Rules fire on everything afterwards

• Recommend a cadence. Weekly for a busy board, monthly for a slow one

Output format

# Pipeline Hygiene · {board}

{n} open leads · median {n} days in stage · {n} need attention

## Diagnosis
| Issue | Count | Worst example |
|-------|-------|---------------|
| Stalled > 30d | 41 | Acme, 94 days in Evaluation |
| Orphaned | 12 | all from Sara (removed 3 weeks ago) |

## Proposed actions
| Action | Count | Detail |
|--------|-------|--------|
| Merge duplicate pairs | 6 | survivor shown per pair |
| Reassign orphans | 12 | → round-robin across {names} |
| Close as lost | 18 | reason required per lead |
| Delete (never a real lead) | 3 | named per lead, confirmed separately |

## Applied
- {n} merged · {n} deleted · {n} reassigned · {n} stages moved · {n} tasks opened
- Held back: {n} ({reason})

## To keep it clean
Saved view proposed · cadence: weekly

Rules and constraints

• Write like a person. Anything a human reads from this skill (a draft, a

note, a task title) is short, plain and specific: short sentences, one ask, the detail that proves you looked. No warm-up lines, no "circling back", no "leverage", no em dashes. A note is three lines of what happened and what is next, not a report. A task title is the action ("Send the 3-seat pricing by Thu"), not a category.

• Never merge without showing what reparents, and never merge on a fuzzy

name match. Duplicates that are actually two people are much worse than duplicates.

• Never close a lead as lost without a reason.
• Every delete and every merge gets its own yes, naming the records it

touches. delete-lead, delete-lead-note, merge-leads and merge-companies cannot be undone from here, so none of them rides a batch approval of the wider plan.

• Delete is for records that were never real. A lost deal is closed; a

duplicate is merged.

• Don't move stages to look busy. A stage change is a claim about reality.
• Bulk cap is 100 per bulk-update-leads and attach-leads-to-event

call. Chunk and report.

• **assign-lead takes user_id; campaigns and departments take

organization_user_id.** Read list-team-members; don't guess.

• Stages are per-pipeline. Always resolve them against the lead's own board.
• Show counts before acting, every time. "This will move 41 leads" is the

sentence that prevents the bad afternoon.

• Never reassign another manager's team's leads without saying so explicitly.

Error handling

• Merge fails → usually a race with another edit, or one side is already

merged. Re-read both with get-lead and re-confirm; never retry blind.

• Assignee no longer a member → list-team-members to confirm removal, then

propose a reassignment target. Report orphans even if you can't fix them.

• Stage not found → you're on the wrong board. get-pipeline-stages for the

lead's pipeline.

• Partial bulk failure → report exactly which ids succeeded and which failed,

then retry only the failures.

• Permission denied → employees can move and update their own leads;

reassignment, deleting leads, the duplicate queue and pipeline structure need leads.manage (managers by default), and company merges need companies.manage. Deliver the plan and name who can run it.

• Note edit or delete refused → it is someone else's note, and the caller is

neither an admin nor a manager over the author's department. Name the author.

• `delete-segment` refused → a live sequence, automation rule or template

variant still targets it, and the refusal names them. Re-scope those first; deleting the filter would widen their audience to every lead.

• Board is empty or brand new → skip the diagnosis theatre. Say the board is

new and suggest lynqu-lead-capture or lynqu-lead-research.

Cross-skill integration

• Duplicates found anywhere in the suite → here for the merge
• Stalled and worth reviving → lynqu-sales-followup
• Stalled because of an incumbent's contract → lynqu-competitors for the

timing trigger, then nurture to that date

• Unscored leads → lynqu-qualify
• Loss reasons accumulate → lynqu-icp (this is where the anti-profile comes from)
• After the sweep → lynqu-pipeline-report, so the manager sees the clean number

Example

"/lynqu pipeline, Sara left last month and I think the board is a mess."

The run: reads three boards · finds 12 leads still assigned to Sara, 41 stalled beyond 30 days (median in-stage is 9), 6 duplicate pairs from the same event captured by two reps, and 88 open leads with no next step · proposes: merge the 6 pairs (survivor shown for each), round-robin Sara's 12 across the two reps in her region, close 18 as lost with reasons, move 9 to nurture with dated triggers, and open a first task on the remaining 61 · applies on approval, then proposes a "stalled > 30d" saved view and a weekly cadence.