| name | dataverse-launchcontrol-business-skills |
| description | Authoritative LaunchControl business policies — launch readiness,
status transitions, escalation, briefings — live as rows in the
Dataverse `skill` table (entity set `skills`) in the LaunchControl
environment. Before answering any LaunchControl business question,
this skill instructs Cowork to load and follow those policies at the
start of the session rather than improvise.
Trigger phrases:
- "is [launch] ready to ship" / "go / no-go on [launch]"
- "what's the readiness for [launch]"
- "this task is blocked, what do I do" / "escalate [task|milestone]"
- "can I move [launch|milestone|task] from [status] to [status]"
- "draft a launch briefing for [launch]"
- "what's our policy on [topic]"
- "what business skills are loaded" / "list LaunchControl policies"
Do NOT use for: ad-hoc CRUD over Dataverse rows (use the
dataverse-launchcontrol-mcp skill) or schema discovery (use
dataverse-launchcontrol-schema).
|
| license | MIT |
| metadata | {"author":"Launch Control","version":"1.4"} |
LaunchControl Business Skills — Policy Loader
LaunchControl ships a curated set of business skills (playbooks,
policies, rules) that live as rows in the Dataverse skill table
(OData entity set: skills). Each row contains the full policy
markdown in the body column. Editing a row in Dataverse changes
agent behavior — the markdown in this plugin is not the source of
truth.
This skill tells Cowork to discover, fetch, and follow those rows
before improvising answers about launch readiness, status transitions,
escalation, or briefings.
When NOT to use
- Ad-hoc record reads / writes →
dataverse-launchcontrol-mcp
- Logical-name discovery /
$expand nav properties →
dataverse-launchcontrol-schema
- Non-LaunchControl topics → other plugins
The skill table
| Column | Type | Purpose |
|---|
skillid | GUID | Primary key |
uniquename | String | Stable identifier (e.g. launch_readiness_checklist) |
name | String | Display name |
description | String | One-line summary — match user intent against this |
body | Memo | Full policy markdown — read and follow verbatim |
OData entity set: skills. Logical name: skill.
Session bootstrap — load the catalog at session start
On the first LaunchControl business question of a session, fetch the
full catalog and cache it for the rest of the session. Do this before
attempting to answer the user's question.
execute
operation: read
query: SELECT TOP 200 skillid, uniquename, name, description FROM skill
If the read returns zero rows, surface the gap explicitly (skill table
empty or user lacks read access) and stop — do not fall back to
embedded policies.
Refresh the cache when:
- The session has been open long enough that policies may have changed.
- The user says something like "the policy changed" or "reload the
skills".
- A policy body references another policy by
uniquename that isn't in
the cache.
Workflow per question
-
Bootstrap — ensure the catalog is loaded (above).
-
Match intent against the cached description column. Prefer the
skill whose description most directly covers the user's intent;
tie-break on name. If two or more skills look plausible, name them
and ask which one applies — do not silently merge.
-
Fetch the body of the chosen skill:
describe
path: skills/<uniquename>
-
Follow the body verbatim. If the body lists valid status
transitions, reject any transition not on the list. If it defines
an escalation chain, follow it in order. If it references other
tables, columns, or nav properties, use the exact names it provides.
-
Cite the skill when answering — e.g. "Per the Launch Readiness
Checklist skill, …" — so the user can trace the answer back to the
Dataverse row.
Hard rules
- Launch readiness is sourced from
lc_launch.lc_risksummary plus the
lc_task rows where lc_isblocked = true. Never hand-tally
milestones, count tasks by status, or compute a percentage in the
prompt and call it "the readiness score". The
dataverse-launchcontrol-mcp skill carries the canonical query pair;
if a skill row body conflicts with that rule, ask the user which to
follow before answering.
- Never invent business policy. If no skill row covers the
question, tell the user the policy isn't defined and ask whether
they want to proceed without one.
- Always operate on unified
lc_* tables, never on the staging
trackers (lc_stg_tracker_a..e) — those were promoted into the
unified model in Episode 3.
- Respect Dataverse permissions. If
execute(operation='read') on
skill returns nothing, the user lacks read access. Tell them — do
not fall back to embedded copies in this skill or in the plugin
bundle.
- Cite, don't paraphrase. Lead with the policy's verdict and cite
the
name + uniquename of the skill row that governed the answer.
Known skills (illustrative — always re-discover via the skill table)
The authoritative list is whatever the session-start read returns. The
list below exists only to orient you:
- Launch Readiness Checklist — go / no-go via
lc_risksummary +
blocked lc_task rows.
- Status Transition Rules — valid transitions per entity
(
lc_launchstatus, lc_milestonestatus, lc_taskstatus).
- Escalation Policy — what to do when a task or milestone is
blocked; includes GitHub-issue cross-check via
lc_GitHubIssueId.
- Launch Readiness Digest — periodic summary format.
- Cowork ↔ Dataverse MCP — Schema-Aware Skill — pointer to
dataverse-launchcontrol-schema for $expand nav lookups.
- Tenant variants (e.g.
contoso-*) — prefer these over the generic
versions when both are present.
Output format
- For policy answers: lead with the verdict / action the policy
prescribes, then cite the skill.
- For "what policies do we have": render a compact table of
name +
description from the cached catalog.
- For multi-step playbooks (escalation, readiness digest): walk the
steps in order, calling the listed tools / queries at each step.
Confirmation gates
- Read-only on the
skill table itself — no confirmations for catalog
/ body fetches.
- Any writes the body instructs you to perform (e.g. updating
lc_taskstatus, creating an lc_statusupdate) follow the
confirmation gates in dataverse-launchcontrol-mcp.
Errors
| Error | Likely cause | Fix |
|---|
skill table not found | Wrong environment, or skills not installed | Confirm env URL; ask admin to run the skill upload script |
| Empty catalog | User lacks read on skill | Surface the permission gap; do not improvise |
body references a table or column that doesn't exist | Skill row is stale | Tell the user; ask whether to proceed without it |
| Multiple matching skills | Overlapping descriptions | Name them, ask which to apply |
Tone
Operational. The audience is a launch manager or PM who wants the
policy-correct answer, not a model's opinion. Lead with what the
policy says to do; explain reasoning only if asked.
