agentleFS
MicrosoftKnown publishermicrosoft/fabric-task-flows/.github/skills/fabric-discover/SKILL.md

fabric-discover

Skill49 starsChanged 6 months agoMITView on GitHub

If the problem statement and project name are already provided (e.g., from auto-chain), use them directly — do NOT re-ask. Otherwise, ask the user for: - Project name — short, creative, descriptive. You MAY suggest examples, but MUST NOT proceed until the user explicitly confirms one. - Problem statement — "What problems does your project need to solve?" The start command runs signal-mapper.py as pre-compute — review its output above. Do NOT run signal-mapper.py again manually. The lexical signal mapper…

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 5 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    DEFAULT first skill — invoke it for ANY new user request that describes a data, analytics, reporting, or integration problem, no matter how …

    line 6
  2. Distinctive:

    If the problem statement and project name are already provided (e.g., from auto-chain), use them directly — do NOT re-ask.

    line 24
  3. Distinctive:

    Read the prompt bundle.

    line 54
  4. Distinctive:

    If the signal mapper generated follow-up questions, use them as templates.

    line 77
  5. Very specific:

    Use --*-source user when the user stated the value, inferred when they deferred to your best guess.

    line 90
  6. Very specific:

    After the block, ask_user with three choices:

    line 100
  7. Very specific:

    Write to _projects/[name]/docs/discovery-brief.md (file is pre-scaffolded — replace full content via edit tool):

    line 107
  8. Very specific:

    Use signal-mapper.py for all signal lookups — do not access registry files directly

    line 128
  9. Very specific:

    Use intake-writer.py to persist 4 V's — do not hand-edit .discovery-intake.json

    line 129
  10. Distinctive:

    Never recommend a final task flow — suggest candidates only

    line 135

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Fabric Discovery

Instructions

Step 1: Collect Intake

If the problem statement and project name are already provided (e.g., from auto-chain), use them directly — do NOT re-ask.

Otherwise, ask the user for:

  • Project name — short, creative, descriptive. You MAY suggest examples, but MUST NOT proceed until the user explicitly confirms one.
  • Problem statement — "What problems does your project need to solve?"

Step 2: Scaffold and Review Signals

python _shared/scripts/run-pipeline.py start "Project Name" --problem "problem statement text"

The start command runs signal-mapper.py as pre-compute — review its output above. Do NOT run signal-mapper.py again manually.

Step 2b: Run the Semantic Capability Mapper

The lexical signal mapper handles vocabulary; the capability mapper handles intent → required Fabric items. It catches what users mean even when they don't use the literal keywords (e.g., "mountain of data to put in data scientists' hands" → Notebook + Lakehouse).

python .github/skills/fabric-discover/scripts/capability-mapper.py \
  --project <project> \
  --text-file _projects/<project>/.problem-statement.txt \
  --emit-prompt-bundle \
  --format json

Exit code handling:

  • Exit 0 — deterministic coverage met. Output already includes required_capabilities and required_items. Move to Step 3.
  • Exit 2 — coverage below threshold. The script wrote _projects/<project>/docs/.capability-llm-prompt.json containing the problem text + candidate intents. You (the agent) MUST:
    1. Read the prompt bundle.
    2. For each candidate_intents[*], decide if the problem text actually expresses that intent. Be conservative — only return high/medium-confidence intents and cite a short quote in rationale.
    3. Persist the response:
      python .github/skills/fabric-discover/scripts/capability-writer.py \
        --project <project> --response-file <path-to-your-json>
      (Or --response-json '{...}' for inline.)
    4. Re-run capability-mapper.py (omit --emit-prompt-bundle) — it now merges your LLM intents with deterministic ones and exits 0.

The output is cached at _projects/<project>/docs/.capability-llm-cache.json keyed by (problem_hash, registry_version) — same inputs produce the same merged result indefinitely.

Step 3: Close 4 V's Gaps (loop until confidence floor met)

For each V not already stated in the problem statement, ask the user one at a time via ask_user. Do NOT proceed to Step 4 with any V still unknown.

VWhat to Assess
VolumeSize per load/day
VelocityBatch / near-real-time / real-time
VarietySources: DBs, files, APIs, streaming
VersatilityLow-code / code-first / mixed

If the signal mapper generated follow-up questions, use them as templates.

Once all four V's are resolved, persist them:

python .github/skills/fabric-discover/scripts/intake-writer.py \
  --project <project> \
  --volume "<value>"       --volume-source user|inferred \
  --velocity "<value>"     --velocity-source user|inferred \
  --variety "<value>"      --variety-source user|inferred \
  --versatility "<value>"  --versatility-source user|inferred

Use --*-source user when the user stated the value, inferred when they deferred to your best guess. If confidence_floor_met is false in the output, loop back and ask more questions.

Step 4: Render the Deterministic Discovery Summary

python _shared/scripts/run-pipeline.py discovery-summary --project <project>

Copy-paste the entire rendered block into your chat response as a fenced code block — the user cannot see tool output directly.

After the block, ask_user with three choices:

  • Confirm — proceed to write the Discovery Brief (Step 5).
  • Correct — capture corrections; re-run intake-writer.py with updated values, then re-render.
  • Restart — abandon and re-collect the problem statement.

Step 5: Produce Discovery Brief

Write to _projects/[name]/docs/discovery-brief.md (file is pre-scaffolded — replace full content via edit tool):

## Problem Statement
[user's exact words]

### Inferred Signals
| Signal | Value | Confidence | Source |

### 4 V's Assessment
| V | Value | Confidence | Source |

### Confirmed with User
- [confirmations/corrections]

### Architectural Judgment Calls
- [max 20 words each]

Constraints

  • Use signal-mapper.py for all signal lookups — do not access registry files directly
  • Use intake-writer.py to persist 4 V's — do not hand-edit .discovery-intake.json
  • Step 4 recap MUST be rendered via run-pipeline.py discovery-summary — never improvise the block
  • Discovery Brief: max 60 lines
  • Signal table cells: max 15 words
  • Architectural Judgment Calls: max 20 words each
  • Integration-first: assume coexistence unless user says migrate
  • Never recommend a final task flow — suggest candidates only

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.