reply-team/jason-ai / skills/business/sdr-operations
sdr-operations
Verifiedoperations/families.yaml
View on GitHub# The SDR business operations contract — family register and fragment schema.
#
# This directory is the source of truth for the contract. Every normative table in this skill,
# and every mapping table in a provider adapter, is generated from these files. Edit here, then
# regenerate. Never edit a generated markdown table.
#
# ---------------------------------------------------------------------------------------------
# FRAGMENT SCHEMA — one file per family, named NN-<slug>.yaml
# ---------------------------------------------------------------------------------------------
#
# family: 8 # must match the NN in the filename
# operations:
# - name: campaign.enroll # entity.verb — two segments, singular snake_case entity,
# # single-word imperative verb. Globally unique.
# core: true # one of the 30 kept in front of the agent at all times
# intent: > # what it is for, in the practitioner's words
# reach: act # read | control | act
# reversibility: compensatable
# approval: confirm_once # auto | confirm_once | confirm_each
# approval_artefact: preview # when the approval is carried by an artefact rather than a prompt
# approval_departs: false # true when the value RAISES above the derivation table below
# protective_floor: false # true for the one legitimate LOWERING: an operation whose
# # non-performance is itself the harm. Requires approval `auto`
# # and a reason, and the build refuses it where the table already
# # derives auto — so the floor is a checked set rather than a
# # phrase somebody remembered to write in a reason.
# approval_reason: > # REQUIRED for either
# before_repeating: > # the observable state to read before running it again
# idempotency_key: required # required | none | conditional
# key_exempt: absolute_valued # REQUIRED on a collection write that takes no key:
# # absolute_valued | terminal_state | unstated. The first two are
# # the shapes a repeat is harmless for; `unstated` is for the rows
# # where the source declares no key and gives no reason, and it is
# # carried visibly rather than justified by invention.
# accepts_collection: false # ONLY on an operation that carries exactly one object.
# # Defaults to true, because absence must read as the dangerous
# # value. A key and a per-item report answer different questions:
# # a key is required on four independent grounds, and only one of
# # them is "accepts a collection". Saying an operation returns no
# # per-item report is therefore a positive assertion about its
# # arity, not an inference from its key.
# per_item_results: required # required | not_applicable | conditional
# cost: none # none | metered
# cost_basis: > # REQUIRED when cost is metered — a basis, never a price
# meter: message # the meter consumed, from the register below. A list where one
# # call buys two different things — `[reveal, enrichment]` — so
# # that both legs stay machine-readable and the preview can
# # project each of them separately, rather than one of them
# # surviving only as prose inside the basis.
# questions: [4] # open-question numbers this operation is unsettled by
#
# A fragment does NOT list the invariants an operation is bound by. Each rule in
# references/invariants.md already names the operations that enforce it, so the binding is
# derived by inverting those lists — one statement, no second copy to drift, and an operation
# named by a rule that does not exist in the contract is a build failure rather than a ghost
# reference nobody notices.
#
# ABSENCE IS NOT A DEFAULT. A missing property reads as the most dangerous value:
# reach `act`, reversibility `irreversible`, approval `confirm_each`, cost `metered`. Nothing is
# left blank because it seemed obvious.
#
# ---------------------------------------------------------------------------------------------
# CONDITIONAL PROPERTIES
# ---------------------------------------------------------------------------------------------
#
# reach, reversibility, approval and cost may each be a scalar OR a conditional block, because
# some operations genuinely change class with state — enrolling into a draft campaign is
# bookkeeping, enrolling into a live one is an act:
#
# reach:
# value: act # what the machine reasons with — ALWAYS the more dangerous side
# conditional: true
# detail: >
# bookkeeping while the campaign is not live, `act` the moment it is; resolved by the
# preview
#
# `value` is the dangerous reading and `detail` is what the agent reads. Every derivation check,
# the approval table and the idempotency rule run against `value`, so a conditional can never
# soften a gate. A conditional without `detail` is an error.
#
# ---------------------------------------------------------------------------------------------
# THE DERIVATION TABLE — approval follows from reversibility x reach
# ---------------------------------------------------------------------------------------------
#
# | reversible | compensatable | irreversible
# -----------|----------------|----------------|---------------------------------------------
# read | auto | auto | auto
# control | auto | auto* | confirm_once
# act | confirm_once+p | confirm_once+p | confirm_each, or confirm_once+p where one call
# | | | is one decision over a set
#
# * confirm_once where the call carries a collection or a pattern, or may overwrite state
# someone else owns.
# +p = with a mandatory preview naming the population, not counting it.
#
# Policy may raise the class and never lower it. Every raise sets approval_departs: true and
# states approval_reason at the operation.
#
# THE PROTECTIVE FLOOR, the catalogue's one checked lowering, never policy's: an operation whose
# non-performance is itself the harm — stopping sending, recording that somebody said stop, opening
# a privacy request, recording an event that already happened outside our control — derives `auto`
# whatever its reversibility. Waiting for permission to stop is the destructive choice. The restart
# direction of every such pair is never `auto`.
#
# ---------------------------------------------------------------------------------------------
# THE TWO STANDING OBLIGATIONS
# ---------------------------------------------------------------------------------------------
#
# 1. An idempotency key is required on every operation with reach `act`, on every write accepting
# a collection, on every write creating a durable object, and on every metered operation —
# including metered reads, because a lost result there costs real money to obtain again.
# Two shapes take no key, because replaying them is harmless by construction: a policy setter
# written as an absolute value ("the ceiling is 400", never "add 200"), and an operation naming a
# terminal state, which a repeat can only arrive at again. They declare which through
# `key_exempt`, and a collection write that declares neither fails the build. The `act` and
# metered grounds are enforced mechanically; "creates a durable object" is not a field and stays
# a reviewer's rule.
# 2. Any operation accepting a collection returns one outcome per item, with the reason. A partial
# success reported as a single verdict is a defect.
#
# ---------------------------------------------------------------------------------------------
# THE METERS
# ---------------------------------------------------------------------------------------------
#
# Named conceptually. Which of them an account holds, and at what granularity it is charged, is an
# account fact and not a contract fact. Every metered operation returns what it actually consumed,
# including zero — "found nothing and was not charged" and "found nothing and was charged anyway"
# are different facts and callers plan differently on each.
contract_version: 2.0.0
# Entity words this contract has retired. They are not aliases and they must never quietly resolve:
# a sentence written today that says `sequence.activate` or `metrics.sequence` is naming something
# that does not exist, and the reader deserves to be told so rather than have the token pass as
# ordinary prose.
#
# This list exists because an entity set derived from the operations alone cannot see them. Both
# words were renamed rather than dropped — `sequence` became `campaign`, `metrics` collapsed into
# `engagement` plus a grouping argument — so neither appears in any fragment, and without this
# every stale reference to them would read as "not an operation reference at all" instead of as a
# defect. It is kept after the migration for the same reason it was needed during it.
retired_entities:
- name: sequence
became: campaign
- name: metrics
became: engagement, with the grouping as an argument rather than a separate operation
# Names this contract deliberately does NOT define. They are listed because the contract sometimes
# has to name an operation in order to say it does not exist — "there is no `consent.check`" is a
# statement worth making, and it would otherwise read to a checker as a reference to a missing
# operation.
#
# Declaring them here does two things a blanket exemption could not. Prose may name them without
# the name-resolution check going blind to a genuine typo in the same file. And a name that appears
# both here and in a fragment is a contradiction the build refuses — so if somebody later adds the
# operation, the sentence claiming it does not exist cannot quietly survive.
declared_absent:
- name: consent.check
instead: consent.prove, or outreach.precheck
because: >
A permission status with no evidence behind it is precisely the answer that cannot be used.
meters:
- id: discovery
consumed_by: searching a database we do not own
typical_basis: per call, or per result returned
- id: reveal
consumed_by: taking ownership of a person found in someone else's database
typical_basis: per person revealed
- id: enrichment
consumed_by: buying named fields for a person or company we own
typical_basis: per field found, per source attempted
- id: verification
consumed_by: checking whether an address or number is reachable
typical_basis: per item checked
- id: monitoring
consumed_by: standing watch on a record for employer change
typical_basis: per record, per period
- id: message
consumed_by: sending, on channels that meter sending
typical_basis: per message, or per message unit
- id: call
consumed_by: dialling
typical_basis: per attempt and per connected minute
- id: profile_read
consumed_by: looking at a person on a channel that meters looking
typical_basis: per profile viewed
- id: signal
consumed_by: observations bought from a feed
typical_basis: per result returned
- id: screening
consumed_by: live screening against an external preference register
typical_basis: per identifier screened
- id: dataset
consumed_by: reputation signals from a paid dataset
typical_basis: per dataset queried
- id: content_generation
consumed_by: drafting a message rather than merging one
typical_basis: per call
- id: request
consumed_by: the installation's own call budget
typical_basis: per call made
- id: invitation
consumed_by: >
asking to connect on a channel that meters invitations, whether or not a note travels with the
ask
typical_basis: per invitation sent
- id: test
consumed_by: >
a deliverability or placement test run through a provider's own seed network, which costs
something to run and reaches nobody real
typical_basis: per test run
- id: registration
consumed_by: >
submitting a sender, a number or a channel for the approval a channel requires before it will
carry anything at all, where the channel charges to register
typical_basis: per registration submitted
# `slug` fixes the generated filenames in every pack:
# this skill -> references/catalog-NN-<slug>.md
# a provider adapter -> references/mapping-NN-<slug>.md
# `area` is the reading order of the guide this contract was drafted from; it groups families for
# a human scanning the index and has no normative force.
families:
- id: 1
slug: discovery
title: Discovery, enrichment and verification
area: People and audience
operations: 12
- id: 2
slug: contacts
title: Contacts, identity and hygiene
area: People and audience
operations: 23
- id: 3
slug: audiences
title: Audience and lists
area: People and audience
operations: 16
- id: 4
slug: import
title: Import
area: People and audience
operations: 5
- id: 5
slug: campaigns
title: Campaign, schedule, windows and limits
area: The programme
operations: 26
- id: 6
slug: steps
title: Steps, content and content policy
area: The programme
operations: 18
- id: 7
slug: messages
title: Messages
area: The programme
operations: 5
- id: 8
slug: enrollment
title: Enrollment
area: The programme
operations: 7
- id: 9
slug: conversations
title: Conversations and activity
area: The conversation, the queue and the day
operations: 9
- id: 10
slug: tasks
title: Task queue and touches
area: The conversation, the queue and the day
operations: 12
- id: 11
slug: meetings
title: Meetings, qualification and handoff
area: The conversation, the queue and the day
operations: 17
- id: 12
slug: signals
title: Signals and triggers
area: The conversation, the queue and the day
operations: 10
- id: 13
slug: inbound
title: Inbound
area: The conversation, the queue and the day
operations: 11
- id: 14
slug: privacy
title: Consent, suppression and privacy
area: Permission and capability
operations: 23
- id: 15
slug: sending
title: Sending capability and deliverability
area: Permission and capability
operations: 34
- id: 16
slug: social
title: Social and messaging channels
area: Permission and capability
operations: 13
- id: 17
slug: oversight
title: Oversight, review and safety
area: The organisation around the work
operations: 32
- id: 18
slug: team
title: 'The team: people, ownership, workload and targets'
area: The organisation around the work
operations: 15
- id: 19
slug: accounts
title: Accounts and buying groups
area: The organisation around the work
operations: 12
- id: 20
slug: measurement
title: Measurement
area: The organisation around the work
operations: 10
- id: 21
slug: introspection
title: Introspection and runtime
area: The organisation around the work
operations: 15
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |