reply-team/jason-ai / skills/business/sdr-operations
sdr-operations
Verifiedoperations/17-oversight.yaml
View on GitHub# Family 17 — Oversight, review and safety (32)
#
# Who may act without asking, what stops for a human to look at, what a human recorded about the
# work afterwards, and what halts the work by itself when nobody is watching.
#
# Schema, the derivation table and the meters: families.yaml.
family: 17
operations:
- name: approval.request
intent: >
"Put something in front of a human and wait" — a message about to go, a bulk write, a piece of
content, a policy change.
reach: control
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: >
the pending queue for this subject and artefact; a request already pending is not a reason to
raise a second one.
idempotency_key: required
per_item_results: required
cost: none
questions: [4]
- name: approval.list
core: true
intent: >
"What is waiting for a human to look at, and how long has it been waiting?"
reach: read
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: nothing at stake
idempotency_key: none
per_item_results: required
cost: none
questions: []
- name: approval.get
intent: >
"Show me this one item exactly as it will go out, with the reasoning behind it"
reach: read
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: nothing at stake
idempotency_key: none
per_item_results: not_applicable
cost: none
questions: []
- name: approval.resolve
core: true
intent: >
"Approve this, or reject it with a reason." Approving is the send.
reach:
value: act
conditional: true
detail: >
`act` on approve, `control` on reject.
reversibility:
value: irreversible
conditional: true
detail: irreversible on approve, reversible on reject.
approval:
value: confirm_each
conditional: true
detail: >
confirm_each on approve, confirm_once on reject — and the resolution is the confirmation when
a human performs it directly.
approval_departs: false
before_repeating: >
the item's state; a resolved item must never resolve twice, and a timed-out approve may already
have sent.
idempotency_key: required
per_item_results: required
cost: none
questions: []
- name: autonomy.get
intent: >
"What may this agent currently do without asking, in this scope, on this channel — and where
does the gate live?"
reach: read
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: >
nothing at stake; re-read at the start of every unattended run.
idempotency_key: none
per_item_results: required
cost: none
questions: []
- name: autonomy.set
intent: >
"Set what runs unattended and what stops" — the gate, the mode, the queue timeout and what a
timeout means, the unattended action cap, the escalation target, and which channels are treated
differently.
reach: control
reversibility: reversible
approval: confirm_once
approval_departs: true
approval_reason: >
raised, because the acts permitted under a loosened setting are not reversible even though the
setting is.
before_repeating: >
the current setting for the scope, and whether an unattended run is in flight under the old one.
idempotency_key: none
key_exempt: absolute_valued
per_item_results: required
cost: none
questions: []
- name: escalation.raise
intent: >
"I cannot or must not proceed; a named human owns this now" — explicitly not an approval,
because nobody is granting permission.
reach: control
reversibility: compensatable
approval: auto
approval_departs: false
before_repeating: >
open escalations for this subject; an open one is a reason to add context, not to raise a
second.
idempotency_key: required
per_item_results: required
cost: none
questions: [4]
- name: escalation.list
intent: >
"What has been handed to a human and not yet picked up, and how old is it?"
reach: read
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: nothing at stake
idempotency_key: none
per_item_results: required
cost: none
questions: []
- name: escalation.resolve
intent: >
"Record how it ended, and whether the work comes back to the agent or stays with the human"
reach: control
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: the escalation's state
idempotency_key: required
per_item_results: required
cost: none
questions: [22]
- name: editorial_review.record
intent: >
"A named human read this machine-drafted message, changed what they changed, and let it go" —
bound to the message actually sent, with the editor, the time and the difference from the draft.
reach: control
reversibility: irreversible
approval: confirm_once
approval_departs: false
before_repeating: >
whether a review is already bound to this sent message; two reviews of one send is an
unresolvable evidence record.
idempotency_key: required
per_item_results: required
cost: none
questions: []
- name: editorial_review.get
intent: >
"Show me the human review behind this message that went out — or tell me plainly that there is
none"
reach: read
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: >
nothing at stake; `none` is a first-class answer and the one an auditor most needs.
idempotency_key: none
per_item_results: required
cost: none
questions: []
- name: review.request
intent: >
"Put a call, a sent message or an unsent draft into someone's review queue" — mine going out for
feedback, or a manager's going to a named reviewer.
reach: control
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: the pending review queue for this artefact
idempotency_key: required
per_item_results: required
cost: none
questions: []
- name: review.list
intent: >
"The review queue: what is waiting, what is done, by person and by period"
reach: read
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: nothing at stake
idempotency_key: none
per_item_results: required
cost: none
questions: []
- name: review.get
intent: >
"One review in full" — the answers, the scorecard version used, who was scored, who scored, and
whether I am entitled to see it.
reach: read
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: >
nothing at stake; a refusal to show a score is a correct answer, not a failure.
idempotency_key: none
per_item_results: not_applicable
cost: none
questions: []
- name: review.score
intent: >
"Score an artefact against a published scorecard" — recording who was scored, who scored, and
which version was used.
reach: control
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: >
existing scores on this artefact for this scored person; a machine-assigned score never
overwrites a human one.
idempotency_key: required
per_item_results: required
cost: none
questions: []
- name: feedback.record
intent: >
"I have actually spoken to them about it" — scored and told-them are different facts and only
one changes behaviour.
reach: control
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: >
whether feedback is already marked given for this review.
idempotency_key: required
per_item_results: required
cost: none
questions: []
- name: review_policy.define
intent: >
"State which completed work gets reviewed, at what sample rate, how far back, and by whom"
reach: control
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: the policy currently in force
idempotency_key: none
accepts_collection: false
per_item_results: not_applicable
cost: none
questions: []
- name: review_policy.get
intent: >
"What is the review policy in force, and what coverage does it actually imply?"
reach: read
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: nothing at stake
idempotency_key: none
per_item_results: not_applicable
cost: none
questions: []
- name: scorecard.define
intent: >
"The questions, their types and their weights, with the arithmetic stated out loud"
reach: control
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: the current draft and its publication state
idempotency_key: none
accepts_collection: false
per_item_results: not_applicable
cost: none
questions: []
- name: scorecard.list
intent: >
"What scorecards exist, which are published, and what is each scoped to?"
reach: read
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: nothing at stake
idempotency_key: none
per_item_results: required
cost: none
questions: []
- name: scorecard.publish
intent: >
"Make this version usable for scoring"
reach: control
reversibility: reversible
approval: confirm_once
approval_departs: true
approval_reason: >
raised, because every score taken afterwards carries this version and past scores must stay
comparable.
before_repeating: the scorecard's publication state and version
idempotency_key: none
accepts_collection: false
per_item_results: not_applicable
cost: none
questions: []
# UNCLEAR: the appendix declares confirm_once against control x reversible, which the derivation
# table produces as auto, and — unlike scorecard.publish, budget.set and stoprule.clear — marks no
# raise and states no justification for it. Transcribed as declared, which is the more dangerous
# side, with the reason drawn from the appendix's own words for the row.
- name: scorecard.retire
intent: >
"Take it out of use for new scoring" — everything already scored keeps its history.
reach: control
reversibility: reversible
approval: confirm_once
approval_departs: true
approval_reason: >
declared confirm_once where control × reversible derives auto, and the contract does not yet
state why. The row's own terms are that it can be published again and everything already
scored keeps its history, so the raise stops at one confirmation. Kept at the declared,
stricter class until the question is settled: an unexplained confirmation costs a sentence,
and lowering a class on a guess costs whatever relied on it.
before_repeating: the scorecard's publication state
idempotency_key: none
accepts_collection: false
per_item_results: not_applicable
cost: none
questions: []
- name: coaching.summarize
intent: >
"Per coach, per period: how much coaching actually happened" — attended, listened, commented,
scored, feedback given.
reach: read
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: nothing at stake
idempotency_key: none
per_item_results: required
cost: none
questions: []
- name: budget.get
intent: >
"What allowance is left on this meter, in this scope, for this period?"
reach: read
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: >
nothing at stake; `unknown` is a real answer, and unknown remaining allowance is not the same as
ample remaining allowance.
idempotency_key: none
per_item_results: required
cost: none
questions: []
# UNCLEAR: "confirm_once, raised whenever it raises a ceiling" reads two ways — one class for the
# whole operation, justified by the case where the ceiling goes up; or confirm_once when it goes up
# and auto when it comes down. Transcribed as the dangerous reading, confirm_once throughout, with
# the appendix's qualifier kept in approval_reason.
- name: budget.set
intent: >
"The ceiling the pre-flight check reads" — a hard stop, not an alert.
reach: control
reversibility: reversible
approval: confirm_once
approval_departs: true
approval_reason: >
raised whenever it raises a ceiling, because the spending it then permits is not reversible.
before_repeating: the ceiling in force and the consumption against it
idempotency_key: none
accepts_collection: false
per_item_results: not_applicable
cost: none
questions: []
- name: stoprule.set
intent: >
"The condition that stops the work by itself" — what is measured, over what window, at what
threshold, and what it halts.
reach: control
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: >
the rules already in force for this scope; two rules on one subject is usually a mistake, not
belt and braces.
idempotency_key: none
accepts_collection: false
per_item_results: not_applicable
cost: none
questions: []
- name: stoprule.list
intent: >
"What stop rules are in force right now, over what scope, and what would each halt?"
reach: read
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: nothing at stake
idempotency_key: none
per_item_results: required
cost: none
questions: []
- name: stoprule.clear
intent: >
"Remove a stop rule"
reach: control
reversibility: reversible
approval: confirm_once
approval_departs: true
approval_reason: >
raised, because removing a safety rule is exactly the write that should never happen by
accident.
before_repeating: >
the rules in force, and whether this rule is currently in a fired state.
idempotency_key: none
key_exempt: unstated
per_item_results: required
cost: none
questions: [25]
- name: stoprule_firing.list
intent: >
"What has fired, when, on what measured value, and what is still stopped because of it?"
reach: read
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: >
nothing at stake; this is the read that tells the next session why sending is off.
idempotency_key: none
per_item_results: required
cost: none
questions: []
- name: stoprule_firing.get
intent: >
"One firing in full" — the rule, the value that tripped it, the operations halted, and every
item left unprocessed.
reach: read
reversibility: reversible
approval: auto
approval_departs: false
before_repeating: nothing at stake
idempotency_key: none
per_item_results: required
cost: none
questions: []
- name: outreach.hold
core: true
intent: >
"Stop everything for this person, this company or this domain, now" — the kill switch.
reach: control
reversibility: compensatable
approval: auto
approval_departs: false
before_repeating: >
no active enrolment, no open task, no pending approval, no un-accepted handoff, no outstanding
meeting proposal and no scheduled message for anyone in scope.
idempotency_key: required
per_item_results: required
cost: none
questions: [10]
- name: outreach.release
intent: >
"Lift a hold"
reach: control
reversibility: compensatable
approval: confirm_once
approval_departs: true
approval_reason: >
raised, because a restart resumes contact with people whose situation may have changed while
they were held.
before_repeating: >
the hold's per-item ledger and each person's current eligibility.
idempotency_key: required
per_item_results: required
cost: none
questions: []
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 |