Management Loop

Management Loop — classify → route → apply → verify → memory

One router + orchestrator over the vault's management skills. It does not replace any of them — it picks the right cluster for the request and chains the handoffs between them. The management/ category has four distinct clusters; this skill is the entry point that keeps them from being mixed up.

management request → classify cluster → route → apply (anti-fluff discipline) → verify → memory
  1. report   (manage-up family)
  2. advice   (persona advisors)
  3. knowledge (methodology knowledge bases)
  4. methodology (qiushi 求是 method)
  └──────────────← gate unmet ←──────────────┘

When to Use

• A management / workplace / leadership request arrives and it is not obvious which of the 28 management skills applies.
• The user asks "which management skill fits", "怎么汇报", "用哪套方法论", or asks for a report AND advice AND methodology in one task.
• A task spans two clusters (e.g. write a weekly report in Amazon style, backed by a thinker's framework) and needs explicit routing + handoff.

The four clusters

The loop

1. Classify the request. Is the deliverable a document (report), advice (persona), knowledge (KB lookup), or method (qiushi)? State the cluster and why. A request that doesn't fit one cluster is usually two — split it, don't blur it.
2. Route. Pick the specific skill(s) within the cluster.
   • report → the concrete report type (weekly / project / proposal / …), always anchored by manage-up-core anti-fluff rules.
   • advice → the persona whose worldview fits (neutral/scissor = god-leader-advisor; long-term/Amazon = bezos-advisor; gray/huawei = renzhengfei-advisor; growth/byteDance = zhangyiming-advisor).
   • knowledge → discipline (cross-company) / individual (thinker) / org (company), per the request's grain.
   • methodology → start with workflows to pick the right 求是 skill sequence; single-method requests go straight to that skill.
3. Apply. Run the selected skill's rules. For any written output, apply the manage-up-core anti-fluff baseline (BLUF, data-anchored, So-What, action-oriented, calibrated language) — even advice and methodology notes benefit from it.
4. Verify. Audit against the source of truth (real data, real principles, real original text — not self-declared).
5. Write memory. Durable decisions, facts, and lessons land in the store the same turn they happen.

Loop back: a gate unmet returns to step 2 (route), not to rewriting blindly.

Rules

1. Classify before you write. "Which cluster" is the first answer, always. Mixing report and advice, or knowledge and methodology, produces mush.
2. Anti-fluff everywhere. manage-up-core's five principles (BLUF / data anchor / So-What / action-oriented / calibrated language) apply to any management output, whatever the cluster.
3. Persona ≠ knowledge base. A persona advisor simulates a leader's judgment; a KB documents a methodology. Don't answer "how does Huawei do X" with the advisor's persona — route to mgmt-org.
4. One grain per lookup. mgmt-discipline (cross-company) / mgmt-individual (thinker) / mgmt-org (company) are three different grains; pick one, don't dump all three.
5. Write memory as you go. Decisions and gotchas land in the durable store the same turn they happen.

Per-round gate (answer at the end of EVERY round)

A round does not close until all four questions are answered explicitly. Do not advance to the next step — or repeat — on an unanswered gate.

1. Success / fail criteria? State the acceptance bar for THIS round: what must be true for the round to count as success vs failure. No criteria = drift, not a round.
2. What should touch / not touch? State the scope boundary: which files / areas / behaviors this round may modify, and which are off-limits.
3. Report criteria? State what the end-of-round report must contain — use the fixed report format below, nothing else as the round summary.
4. Self-review rounds + next step? State how many critic rounds you will run before accepting, and whether the next step auto-starts and repeats.

Round report (fixed format)

At the end of every round, emit exactly this block — nothing else as the round's summary:

success criteria: <restate this round's original criteria verbatim>
criteria status: <one line per criterion: met / not met / partial, with evidence>
success confidence: <0-10>
failure confidence: <0-10>
touch: <files/areas modified>
not touch: <files/areas deliberately left alone>
next: <single next action>
self review status: <critic rounds run, blocking issues remaining>
next step status: <auto-start | wait-for-user | done>

The two confidence scores (0–10) are the agent's own estimate of how likely the round succeeded and how likely it failed — they are NOT a verdict. The human decides success/failure from the criteria status, not from the scores.

Termination

The loop ends when the request's completion audit passes against the real source — report is data-anchored and fluff-free, advice is persona-consistent, knowledge is correctly grained, methodology is applied — and the final round report shows all criteria met with next step status: done. The agent never declares the loop a success itself; the human does, from the criteria status.

When NOT to use

• A single skill obviously covers the request — don't route for the sake of routing.
• Non-management domains (code, content writing, math) — use their own skills.
• A task already inside one cluster with a clear single answer.