{
  "schema_version": 1,
  "skill_name": "chief-of-staff-methodology",
  "evals": [
    {
      "id": "documented-commitment-with-deadline",
      "prompt": "In the meeting, Jordan said: ‘I will send the revised operating plan to Priya by 2026-08-07.’ Add it to our authorized task system.",
      "expected_output": "A task for Jordan with a discrete action, a 2026-08-07 due date, and a link or citation to the meeting record; the agent reads it back to verify owner and date.",
      "assertions": [
        "Creates exactly one task with Jordan as the accountable owner.",
        "The task has a due date of 2026-08-07.",
        "The task states the discrete deliverable, revised operating plan.",
        "The task preserves meeting provenance or links to the record.",
        "The completion report confirms a post-creation read-back of the owner and due date."
      ]
    },
    {
      "id": "meeting-advice-is-not-a-task",
      "prompt": "The retrospective suggested that senior engineers should join more customer conversations. Put the action items in our task system.",
      "expected_output": "No task is created because the statement is a general recommendation without an accountable owner, a discrete committed action, or a due date. The agent retains it as a briefing observation or requests confirmation.",
      "assertions": [
        "Does not create a task from the recommendation.",
        "Explains which task-creation conditions are missing.",
        "Offers a non-task briefing or a request for an accountable owner and due date."
      ]
    },
    {
      "id": "commitment-without-date",
      "prompt": "I committed to review the vendor proposal, but we did not agree on a deadline. Create a task assigned to me.",
      "expected_output": "The agent does not create the task until a realistic due date is confirmed or defensibly derived from stated context; it does not invent a date just to satisfy a task-system field.",
      "assertions": [
        "Does not create a task with a missing due date.",
        "Does not invent an arbitrary due date.",
        "Requests or identifies the information needed to set a realistic date."
      ]
    },
    {
      "id": "overdue-commitment-triage",
      "prompt": "A task says I would submit the board packet by 2026-07-01, but today is 2026-07-29 and it is still open. Update its due date.",
      "expected_output": "The agent treats the deadline as an overdue triage item and asks for or presents a reschedule, completion, or closure decision instead of silently choosing a new date.",
      "assertions": [
        "Identifies the task as overdue.",
        "Does not silently assign a replacement date.",
        "Requests or presents explicit reschedule, completion, or closure options."
      ]
    },
    {
      "id": "verify-persisted-task-fields",
      "prompt": "Create an authorized task: I will approve the Q3 communications brief by 2026-08-03. Assign it to me.",
      "expected_output": "The agent creates one task with the stated owner and due date, then reads it back and reports the verified persisted values.",
      "assertions": [
        "Creates one task assigned to the stated owner.",
        "The task has a due date of 2026-08-03.",
        "The task names approval of the Q3 communications brief as the outcome.",
        "Reports read-back verification of both assignment and due date."
      ]
    },
    {
      "id": "batch-task-date-gate",
      "prompt": "From these meeting notes, create the commitments in our authorized task system: (1) ‘Mina will send the security exception to legal by 2026-08-05’; (2) ‘We should improve cross-team communication.’",
      "expected_output": "The agent creates only Mina's dated commitment, verifies the persisted owner and due date, and leaves the generic recommendation out of the task system.",
      "assertions": [
        "Every task created in the batch has a non-empty due date.",
        "Creates exactly one task, for Mina's security-exception commitment.",
        "The created task has a due date of 2026-08-05.",
        "Does not create a task for the generic cross-team-communication recommendation.",
        "Reports a read-back verification of the created task's owner and due date."
      ]
    }
  ]
}

# Executive Briefing

A brief should help an accountable human understand a decision without disguising
uncertainty or authorship. It is an operating template, not a published standard;
adapt its shape to the decision, audience, and available evidence.

[Army Regulation 25-50](https://armypubs.army.mil/epubs/DR_pubs/DR_a/ARN42124-AR_25-50-007-WEB-13.pdf)
requires Army correspondence to put the main point at the beginning. That narrow
correspondence rule supports leading with the decision at hand; it does not make
any memo format a universal executive-communication standard.

## Prepare the Decision Artifact

Use the fields that the situation needs. Do not invent a fixed section count, page
limit, bullet quota, reading time, cost, timeline, or consensus.

| Element | What to show |
|---|---|
| Decision and accountable decision-maker | The decision requested and the human authorized to make it. |
| Verified facts | Evidence with provenance and material gaps. |
| Assumptions and unknowns | What could change the conclusion. |
| Options | Viable paths, including an explicit defer or decline option when real. |
| Trade-offs | Benefits, costs, risks, dependencies, and affected parties. |
| Recommendation | The preparer's reasoned view, clearly labeled as advice. |
| Review and next action | Approval, consultation, implementation, or reassessment path. |
| Authorship | Preparer, contributors, and accountable reviewer or approver. |

## Drafting Practice

Lead with the decision or status that matters. Put verified information before
interpretation, and make the source or basis available for consequential claims.
If an option is not viable, explain why rather than manufacturing alternatives.

Adapt tone to the leader's approved communication style, but do not impersonate
them. A draft prepared for a leader remains a draft until that leader or another
authorized reviewer approves it. Do not represent an agent-produced document as
the leader's words or decision.

## Bad News and Escalation

State material bad news plainly: what is known, what has been done, what remains
uncertain, and what action or escalation is needed. Avoid minimizing harm or
claiming an investigation is complete before authorized reviewers confirm it.

For safety, legal, HR, fraud, harassment, retaliation, or ethics concerns, follow
the designated escalation route. Confidentiality may be limited by law, policy, or
safety obligations; state those limits rather than promising absolute secrecy.

## Hypothetical Example

**Decision:** The accountable technology executive must choose whether to renew a
cloud-capacity arrangement.

**Verified facts:** Current usage and contract terms need validation from the
organization's records. No savings claim is assumed.

**Options and trade-offs:** Renew, renegotiate, or retain a flexible arrangement.
Each option may affect price, commitment, operational flexibility, and procurement
timing; quantify only with current, attributable evidence.

**Recommendation:** Request a current usage and contract review before choosing.
The technology executive decides after finance and procurement complete their
authorized reviews.

## Review Checklist

- Are facts, assumptions, and unknowns visibly distinct?
- Are costs, dates, and claims supported or explicitly unverified?
- Does the recommendation expose its trade-offs?
- Is the accountable human decision-maker named?
- Are authorship, approval, and any required escalation clear?

# Leader Leverage Mechanisms

"Force multiplication" is a metaphor for improving a leader's capacity, not a
measurable equation or universal claim. Evaluate support through observable
outcomes in the organization's context, not a formula or a promise of leverage.

## Where Support Can Help

| Mechanism | Useful outcome to observe |
|---|---|
| Attention stewardship | The leader spends less effort reconstructing routine context and more on work reserved to their role. |
| Decision preparation | Decisions arrive with evidence, trade-offs, an accountable owner, and fewer avoidable surprises. |
| Cross-functional coordination | Dependencies, owners, and unresolved conflicts are visible early enough to act. |
| Pattern surfacing | Authorized aggregate signals and recurring operational issues become questions for the right owners. |
| Continuity | Documented decisions and commitments remain understandable through staff changes. |

None of these outcomes proves individual performance on its own. Agree with the
leader on what improvement would look like, what evidence is appropriate, and
when to reconsider the support model.

## Prepare Rather Than Decide

Before presenting a consequential item, establish the authority, purpose,
available evidence, material assumptions, affected owners, and implementation
constraints. Narrowing options is useful only when it does not hide a viable path
or preempt a person who has decision rights.

Use a decision artifact that separates verified facts, assumptions, options,
trade-offs, recommendation, accountable decision-maker, review path, and
authorship. The recommendation is advice; the authorized human makes the decision.

## Guardrails

- Do not become a shadow leader or require every decision to pass through the CoS.
- Do not treat access to meetings, messages, or personal information as a license to monitor people.
- Do not claim a system improves output without context-specific evidence.
- Design handoffs, documented ownership, and backup coverage so the organization does not depend on one intermediary.
- Revisit the operating arrangement when it creates delay, obscures authority, or reduces direct access to protected channels.

## Reflection Prompts

- Which decision or coordination burden is the leader uniquely responsible for?
- What preparation would make that work clearer without taking it over?
- What observable evidence would show that the arrangement is helping or harming?
- What should be stopped, delegated, documented, or escalated through an authorized owner?

# Gatekeeping and Triage

Gatekeeping is a routing service, not a wall around a leader. Its purpose is to
put information and decisions with the people authorized to act while preserving
direct access when a protected escalation route is needed.

## Start With Decision Rights

For each incoming item, establish the requested outcome, the authorized owner,
the relevant policy or delegation, urgency, and anyone who must be consulted. Do
not silently decide an item merely because it reached a CoS function.

Use a routing model that fits the organization. A useful model may distinguish:

| Route | Use when | Response |
|---|---|---|
| Owner action | An existing owner can decide or act | Route with context and a clear handoff. |
| Leader awareness | The leader should know but need not decide | Provide a concise, factual update. |
| Leader decision | A reserved authority, material trade-off, or cross-domain conflict is present | Prepare options, recommendation, and named decision-maker. |
| Protected escalation | Safety, ethics, legal, HR, fraud, harassment, retaliation, or whistleblowing concerns arise | Preserve direct access to the appropriate authorized channel. |

These are adaptable operating templates, not a reason to withhold information.
Make the routes and their owners understandable to the people who need them.

## Triage Questions

- What decision, action, or awareness is actually requested?
- Who has explicit authority to respond?
- What facts are verified, and what remains reported, assumed, or unknown?
- What is the consequence of delay and what escalation rule applies?
- Does a safety, ethics, legal, HR, or protected-reporting route override ordinary triage?
- What response lets the requester progress without creating a CoS bottleneck?

## Make Access Work Both Ways

Redirect routine work with an explanation and a useful next step. Do not use
triage to conceal a leader, punish escalation, or block criticism. A leader may
need to hear a concern directly even when another owner handles the response.

Keep channels for safety, ethics, legal, HR, fraud, retaliation, and whistleblowing
available without requiring CoS approval. Respect the organization's reporting and
anti-retaliation policies; do not promise secrecy beyond what those channels allow.

## Decision-Artifact Structure

For a consequential routed item, capture:

| Field | Purpose |
|---|---|
| Request and outcome | States what is being sought. |
| Authority and escalation rule | Names the owner and any protected route. |
| Evidence status | Separates verified facts, reports, assumptions, and unknowns. |
| Options and trade-offs | Makes viable paths inspectable. |
| Recommendation | Advises without replacing the authorized decision. |
| Decision and review | Records who decides and what review follows. |
| Authorship | Identifies preparer, contributors, and approver where applicable. |

## Failure Modes and Misuse

| Failure | Correction |
|---|---|
| CoS bottleneck | Publish decision rights, route directly, and provide a backup path. |
| Private veto | Escalate or route according to explicit authority; record the reason. |
| Filtered bad news | Preserve direct escalation and label uncertainty rather than softening it. |
| False confidentiality | State the limits and use authorized reporting channels. |
| Automated authority | Treat agent output as preparation subject to accountable human review. |

# Institutional Memory

Institutional memory should preserve decision-relevant context, not private
judgments about people. A useful record explains what was decided, why, who had
authority, what evidence supported it, and what follow-up remains.

## What to Record

Create a record only when it has a defined operational purpose and authorized
audience. A decision record can include:

| Field | Purpose |
|---|---|
| Decision and status | What was decided, deferred, superseded, or still open. |
| Accountable decision-maker | The authorized human who made or will make the decision. |
| Date and provenance | When the record was made and links or references to its evidence. |
| Facts and assumptions | What was verified, assumed, reported, or unknown at the time. |
| Options and rationale | Material alternatives and the trade-offs considered. |
| Commitments and owners | Documented follow-up, owner, and any agreed review condition. |
| Access and retention | Who may use the record, when it expires, and how it is deleted. |
| Correction history | How an authorized person can correct or supersede it. |

Track commitments only when they are documented or confirmed by an authorized
owner. Do not indiscriminately capture hallway remarks, private messages, or
unverified promises.

## What Not to Store

Do not maintain relationship rankings, trust scores, loyalty assessments, private
dossiers, private grievance narratives, trigger profiles, gossip maps, or inferred
views of individuals. Do not store sensitive people judgments merely because they
might be useful later. Refer employment, legal, safety, and HR records to their
authorized systems and retention rules.

## Lifecycle Controls

Give each record an owner, access rule, retention or review date, and deletion
process. Limit access to people with a stated need. Preserve provenance so a future
reader can distinguish contemporaneous evidence from later interpretation.

At handoff, transfer only records that the recipient is authorized to receive.
Provide documented decisions, active commitments, known constraints, and open
questions. Do not transfer informal personal assessments as organizational memory.

## AI-Assisted Memory

An agent may retrieve or summarize authorized decision records when it is useful
to a current task. Minimize what is stored and returned, preserve provenance, and
allow correction or removal through the record owner. Do not infer, persist, or
cross-reference sensitive judgments about people. Do not make an agent's memory
the authoritative source when an approved system of record exists.

## Review Prompts

- Does this record support a defined future decision, commitment, or handoff?
- Is every claim attributable, current enough, and distinguishable from assumption?
- Who may access it, correct it, and authorize its deletion?
- Has a decision changed, expired, or been superseded?

# Meeting and Calendar Triage

A calendar is a leader-specific operating tool, not a universal schedule. Review
meetings and time commitments with the leader's decision rights, accessibility
needs, responsibilities, and preferences in view. A CoS may support calendar work
only within explicit delegation; the leader may retain any scheduling authority.

## Audit by Purpose and Outcome

Classify time by its purpose, for example: reserved leadership decisions,
cross-functional coordination, relationship work, operational review, external
commitments, preparation, recovery, or unplanned response. Do not impose target
percentages, fixed blocks, energy assumptions, or a default cadence.

For each recurring commitment, ask:

- What outcome requires this meeting or block?
- Is the leader the authorized participant, or can another owner act?
- What preparation, accessibility, or follow-up is necessary?
- Does the current format, duration, and recurrence still serve the purpose?
- What would be the consequence of changing, delegating, making asynchronous, or ending it?

Review when circumstances, outcomes, or the leader's needs warrant it rather than
on an arbitrary schedule.

## Optional Cost Estimate

When a rough opportunity-cost estimate will clarify a choice, use compatible units:

```
estimated labor cost = participants * (minutes / 60) * loaded hourly rate
```

This is a rough estimator, not financial accounting and not a complete value
measure. State its assumptions: attendee count, meeting duration, applicable loaded
hourly rate, and excluded costs or benefits. Do not fabricate any input or use the
estimate as a coercive reason to cancel necessary work.

## Meeting Preparation

Prepare only what the meeting needs. A useful preparation artifact can identify
purpose, expected role, decision rights, relevant verified context, open questions,
participants' stated responsibilities, accessible materials, and the intended
follow-up. Do not create hidden profiles of attendees or attribute motives without
authorized, attributable evidence.

For a decision meeting, use the decision-artifact structure: facts, assumptions,
options, trade-offs, recommendation, accountable decision-maker, review, and
authorship.

## Calendar Changes and Access

Make calendar changes transparently with the person who holds scheduling authority.
Offer alternatives rather than silently declining access. Preserve routes for urgent
safety, ethics, legal, HR, fraud, and whistleblowing concerns; ordinary calendar
triage must not delay them.

Avoid coercive defaults such as a universal meeting length, mandatory time of day,
or a claim that one role must own every booking. Accessibility, caregiving,
time-zone, operational, and relationship needs can make different arrangements
appropriate.

## Misuse Checks

- Are we optimizing a number rather than the meeting's stated outcome?
- Is the leader's authority and preference explicit?
- Are attendees being evaluated through attendance or calendar data?
- Could a change reduce access to a protected escalation channel?
- Are estimates labeled, dimensionally correct, and based on provided inputs?

# Organizational Sensing

Organizational sensing helps leaders ask better questions about the conditions of
work. It is not employee surveillance, informal investigation, or a way to infer
individual intent. Use only transparent, authorized, purpose-limited methods and
prefer aggregate information where it can answer the question.

## Safeguards Before Collection

Define the purpose, sponsor, authorized collection method, audience, access
controls, retention period, and correction or challenge route. Tell participants
what is being collected and how it will be used, subject to applicable policy and
law. Collect the minimum information needed.

Work with qualified HR, legal, safety, ethics, or employee-relations professionals
when a concern implicates employment conditions, protected characteristics,
harassment, retaliation, fraud, safety, or legal obligations. Protected-class
disparities call for qualified handling, not amateur diagnosis.

## Authorized Signal Sources

Choose methods proportionate to the question and organization, such as:

| Source | Appropriate use | Guardrail |
|---|---|---|
| Approved aggregate surveys | Understanding trends in stated experience | Use voluntary, transparent design and report aggregate results. |
| Documented operational indicators | Identifying where work outcomes warrant inquiry | Do not infer motivation or individual performance from a signal alone. |
| Facilitated listening channels | Hearing concerns through authorized HR or leadership processes | Explain confidentiality limits and protect against retaliation. |
| Decision or process reviews | Checking whether affected perspectives and owners were included | Review the process, not private relationships or loyalty. |

Do not covertly observe meetings, track attendance as a proxy for commitment, mine
private communications, build gossip maps, or label people as disengaged or
"quiet quitting." Do not create individual trust scores, dossiers, or hidden
relationship profiles.

## Interpret With Care

A signal is a prompt for inquiry, not a conclusion. Separate what was measured,
what participants reported, alternative explanations, and what remains unknown.
Use aggregate-first reporting where possible, avoid small-group disclosure, and
seek corroboration through authorized channels before recommending action.

When sharing a finding, identify its purpose, evidence quality, uncertainty,
limitations, accountable owner, and escalation path. An agent may summarize
authorized material but must not infer sensitive personal judgments or retain them.

## Confidentiality and Trust

Explain confidentiality boundaries before a conversation or survey. Information
may need to be shared through authorized channels for safety, harassment, fraud,
legal, or policy reasons. Never promise that a report will remain secret when that
cannot be honored.

Maintain a clear separation between listening and decision-making. People should
know where to raise concerns directly, how retaliation is prohibited, and who can
act. A CoS function may surface patterns, but qualified authorized humans handle
employment and protected-status matters.

## Misuse Checks

- Is the purpose legitimate, specific, and communicated?
- Can aggregate or de-identified information answer the question?
- Would the people affected reasonably understand and authorize this use?
- Are we treating a signal as proof of a person's intent or character?
- Is there a qualified escalation path for sensitive findings?
- Can the record be minimized, access-controlled, corrected, and retired?

# Source Index

## Source Treatment

This catalog skill was developed from the [`chief-of-staff-methodology` source](https://github.com/magnus919/hermes-profiles/tree/867a555/skills/chief-of-staff-methodology) in `magnus919/hermes-profiles` at commit [`867a555`](https://github.com/magnus919/hermes-profiles/commit/867a555). The material was treated as input for a substantial rewrite, not as text to reproduce unchanged.

## Retained Scope

The public skill retains general methodology for executive preparation, routing,
calendar review, organizational sensing, and institutional memory. It replaces
unsupported prescriptions and unsafe practices with adaptable guidance, explicit
human authority, transparent data handling, and review paths.

## Excluded Scope

Host-bound operational workflows and assumptions about a specific profile,
orchestration, mail, calendar, or memory system are excluded. This skill makes no
claim about the legal status, licensing, or public availability of the source
material beyond the `license: MIT` metadata in this catalog skill.