hollandkevint/data-product-operator / skills/team-loop
team-loop
UnverifiedFormerly dpm_os. Initialize a Team Loop workspace, a markdown workspace for one data product operator to move one data product decision from discovery to controlled release with three manual gates, a weekly decision packet, and an append-only decision log. Detects greenfield, migration, or active-repo mode, asks only for missing load-bearing context, copies the deterministic scaffold without overwriting, populates fields with provenance tags, runs the validator, and hands off the first three actions. Use when invoked as `/team-loop` (formerly `/dpm-os`) or when asked to set up a Team Loop in a directory.
team-loop — initializer (formerly dpm_os)
Scaffolds and initializes a team loop workspace in the current working directory. Structure is deterministic (copied from scaffold/). Reasoning is adaptive (loaded from prompts/ per phase).
| Layer | Where | Why |
|---|---|---|
| Static structure: manual, index, context files, gates, templates, schemas, validator | scaffold/ | Same every time. Never regenerated. |
| Adaptive reasoning: mode detection, migration, interview, post-scaffold | prompts/ | Depends on the directory and the operator's answers. |
| Orchestration | this file | Glue. |
When to invoke
- The operator runs
/team-loopor asks to set up a team loop here. - Do not invoke for routine work after init. The seeded
CLAUDE.mdin the instance owns that.
Workflow
1. Detect mode
Load prompts/mode-detection.md. Decide greenfield (empty), migration (product artifacts present), or active-repo (a working software repo). Announce the mode in one line. In active-repo mode do not proceed without the operator's confirmation.
2. Migrate if needed
Load prompts/migration.md. Copy pre-existing artifacts into source/. Never move. Record cross-document conflicts for the post-scaffold contradictions block.
3. Interview
Load prompts/interview.md. Ask the four batches, or only the gaps the source artifacts leave open. Confirm back what you heard before scaffolding.
4. Copy the scaffold, never overwrite
Copy every file and folder from scaffold/ into the current directory, including .claude/, .gitignore, and scripts/. Run exactly this form, which carries dotfiles and never overwrites:
cp -Rn "<skill-dir>/scaffold/." .
Plain cp -R overwrites collisions; do not use it. On macOS, cp -n exits 1 when it skipped an existing file; that is a collision report, not a failure. Then apply these rules:
- A path that already existed at the destination was skipped by
-n. List each such path as a collision. .claude/settings.jsonalready present: merge the scaffold's onePostToolUsehook entry into the existinghooks.PostToolUselist. Do not replace the file.- The current directory is the project root. Do not create a nested subfolder.
- Do not modify files under
scaffold/at the source.
Verify the copy with python3 scripts/validate.py .. Any required file missing finding means the copy failed; fix it before continuing. Other findings are expected until step 5 fills the context files.
5. Populate with provenance
Walk the copied context/ files and fill each section from the interview or the source artifacts, per the table in prompts/interview.md. Each context file you fill carries a tag under its ## Provenance heading, from the enum in decisions/_SCHEMA.md or the literal Unknown. The validator in step 6 checks this. Fill the ## Current decision block in INDEX.md; when no decision exists yet, write the first decision the operator named and mark its evidence gap.
6. Validate
Run python3 scripts/validate.py . from the instance root. Fix every finding before continuing.
7. Commit
If the directory is inside a git work tree, stage the scaffolded files and commit once: feat: initialize team loop. Otherwise git init here first. Never push. The operator controls publication.
8. Hand off
Load prompts/post-scaffold.md. Lead with the three first actions, then contradictions, then two or three gaps, then one paragraph on what was built.
Skill surface
This initializer does not bundle analysis skills. When the data-product-operator skill library is installed under .claude/skills/ at project or user scope, these resolve by name. When it is absent, the names below will not resolve; install it first (see install).
| Skill | Reach for it when |
|---|---|
data-product-thinking | Framing the problem before the data |
data-consumer-discovery | Finding the first user and the decision they make |
data-product-validation | Testing demand before building |
research-synthesis-data | Turning interviews into evidence rows |
data-team-positioning | Deciding whether the team is a service or a product team |
metrics-definition | Writing the outcome metric with grain and window |
data-quality-assessment | Filling the quality assertions table |
data-pipeline-quality | Testing the pipeline behind a claim |
data-model-design | Reviewing the schema behind a contract |
stakeholder-alignment | Filling decision rights and bringing owners in early |
data-team-operating-model | Setting cadence and handoffs |
ethical-risk-assessment | The fifth risk in the assumptions list |
healthcare-data-domain | Any regulated health data context |
data-storytelling | Writing the packet's recommendation |
dashboards-to-decisions | Turning a dashboard request into a decision |
grill-me-data | Pressure-testing a packet before the gate |
arbitrage-audit-data | Finding where the data creates an edge |
data-product-retro | Looking back after a launch or an incident, with actions tied to log rows |
analytics-agent | Planning or testing chat over the data product |
metric-glossary | Settling metric names, owners and grains |
launch-and-adoption | Planning the launch for a named consumer |
shipping-bridge | Mapping the controlled-release gate to release practices |
Loops that read this workspace
Each loop follows the shared loop pattern. A loop reads the decision log and the files named in its skill. It reports and never edits the workspace.
| Loop | Reach for it when |
|---|---|
decision-follow-up | Weekly, after the packet: which decisions are past their review date |
weekly-metric-review | Weekly, before the team reads its numbers: which moves have no cause |
data-quality-watch | Daily, after the load: whether the data can be trusted today |
agent-eval-regression | After a model, prompt or schema change: whether an agent got worse |
strategy-review | Quarterly: what to keep, change or stop |
Anti-patterns
- Regenerating scaffold content instead of copying it.
- Overwriting a file the operator already had.
- Skipping the validator.
- Inventing contradictions. If none were found, say so.
- Pushing.
Files
SKILL.md— this fileprompts/mode-detection.md,prompts/migration.md,prompts/interview.md,prompts/post-scaffold.mdscaffold/— the instance, copied as-is
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: 196k | 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: 196k | 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: 196k | Updated | Claude CodeCodex | Verified | Install |