Files
agent-prod/skills/lead-radar/SKILL.md
T

94 lines
5.8 KiB
Markdown
Raw Normal View History

2026-11-01 15:59:55 +01:00
---
name: lead-radar
description: |
Sascha's leadership radar: read the IT cockpit through one lens, turn the situation into
numbered decision proposals, execute them in Basecamp/Zammad after one batch approval.
Modes: tag (morning Lagebild, Zammad + 👥 IT), freitag (prepare the Friday close),
meins (weekly review of Sascha's own 🔒 Meins to-dos), person <slug> (prepare a 1:1),
portfolio (monthly review of Sascha's projects). Also answers "wem delegiere ich X"
via person mode.
Triggers: "Lagebild", "Freitags-Abschluss vorbereiten", "Wochenreview", "1:1 vorbereiten",
"Portfolio-Review".
compatibility: opencode, claude-code
---
# lead-radar
Guiding question of every run: *What must Sascha decide now, and whom does he give it to?* One run = one mode → one Lagebild → one numbered proposal list → one batch approval → execution.
Invocation: `/lead-radar <tag|freitag|meins|person <slug>|portfolio>`. Without a mode, pick from the user's words via the trigger phrases above; still unclear → ask with the five modes as options.
## Sources of truth
Read before acting; these hold the facts, this skill holds only behaviour:
- `/home/sascha.koenig/p/PM/.okf/processes/it-basecamp-regeln.md`: task rules, the six 👥 IT lists, 30-day rule, Zammad boundary, **approval protocol** (binding for every write)
- `/home/sascha.koenig/p/PM/.okf/systems/basecamp.md`: project, list and folder IDs
- `/home/sascha.koenig/p/PM/.okf/organization/*.md`: people (role, skills, Basecamp person ID, Zammad login)
- `/home/sascha.koenig/p/PM/tools/it-cockpit/README.md`: the data layer and what each scope returns
An ID that fails: resolve it by name, carry on, and add the mismatch to the Drift block.
Load the `basecamp` skill before any Basecamp CLI call; check flags with `basecamp <cmd> --agent --help`, they evolve. For German text that colleagues read (comments, notes), follow `sascha-internal-communication-style`.
## Steps
### 1. Pull the data
From `/home/sascha.koenig/p/PM`:
```bash
python3 tools/it-cockpit/cockpit.py --json --scope <mode> [--person <slug>]
```
A run takes 20–90 s. Save the JSON to the scratchpad and work from that file. Every output carries `drift` and `errors`; an entry under `errors` (e.g. Zammad without `ZAMMAD_TOKEN`) means that source is blind for this run: say so in one line at the top of the Lagebild and continue with the rest. Use live `basecamp` calls only for detail questions the cockpit does not answer (comments under an item, a card's history).
**Done when** the JSON is on disk and you know which sources are blind.
### 2. Read the situation through the mode's lens
Open the mode's reference and follow it; it says which keys to read, how to judge them, and which proposals each finding yields:
| Mode | Reference | Persists |
|---|---|---|
| `tag` | [references/tag.md](references/tag.md) | nothing |
| `freitag` | [references/freitag.md](references/freitag.md) | nothing |
| `meins` | [references/meins.md](references/meins.md) | doc in 🔒 Meins → `Wochenreviews` |
| `person <slug>` | [references/person.md](references/person.md) | nothing, by hard rule |
| `portfolio` | [references/portfolio.md](references/portfolio.md) | doc in 🔒 Meins → `02 Projekte & Initiativen` |
**Done when** every item in the mode's buckets has either a proposal or a one-line reason why it needs none.
### 3. Present: Lagebild, proposals, drift
Answer in German, in this order:
1. **Lagebild**: the mode's template from its reference. Judgement first, lists second; every item with its link.
2. **Vorschläge**: one numbered list across the whole run. Each entry: number, object (title + link), action, new value (owner / date / list), and the comment text that will go on the object. Use only the actions the approval protocol allows; anything outside it (archive a project, create a list, ping someone) becomes a point for Sascha himself, marked `Für dich:` and left unnumbered. A stalled project or one that is really just an idea becomes `Für dich: /lead-backlog <Projekt>`.
2026-11-01 15:59:55 +01:00
3. **⚠️ Drift** (only when `drift` is non-empty or step 1 hit an ID mismatch): one line per finding, followed by numbered proposals in the same list that fix the OKF side (e.g. update an ID row in `systems/basecamp.md`, add a minimal profile via `lead-profile`). Findings whose fix is a team decision (`liste_ausserhalb_regel`) go to the Friday close as `Für dich:` points.
End with the approval question: "Freigabe für alle, Abwahl per Nummer, oder einzelne Punkte ändern?"
**Done when** the user has answered explicitly. Silence or a follow-up question is not approval; answer it and ask again.
### 4. Execute the approved items
Follow [references/execute.md](references/execute.md): one command per change, the short "why" comment on every changed object, read-back afterwards.
**Done when** every approved item is either verified in Basecamp/Zammad/OKF or reported as failed with the exact command and error.
### 5. Persist and report
For `meins` and `portfolio` write the document as the mode's reference describes, reflecting what was actually executed. Other modes write nothing.
Report in German: executed items with links, failed items with error, document URL (if any), open `Für dich:` points.
**Done when** the report is out and, for `meins`/`portfolio`, the document is readable in Basecamp.
## Guardrails
- Every write waits for the batch approval; the approval covers exactly the numbered items shown.
- Proposals carry their reason; a bare "Termin verschieben" is incomplete.
- OKF holds goals, context, decisions and links; task state lives in Basecamp (spec §3.2). OKF fixes go through the same approval and end with the `validate` skill on the bundle.
- Process KPIs only: counts per list or per process, never per-person rankings.
- Owner reminders belong to automation A3; the skill's output to colleagues is the comment on the object, nothing else.