skinnyandbald/ceos-scorecard-autopull

ceos-scorecard-autopull

Enhanced weekly scorecard entry that synthesizes data from L10 meetings and any available MCP sources (CRM, calendar, email, filesystem). Walks through each metric one at a time with a pre-populated suggestion, letting the user confirm or tweak before proceeding to the next metric.

This skill wraps ceos-scorecard -- it does NOT replace it. After interactive confirmation, it writes the file using the same format as ceos-scorecard Mode: Log Weekly.

When to Use

Use this anytime you want to log scorecard numbers with auto-populated suggestions instead of entering them manually.

Context

Finding the CEOS Repository

Search upward from the current directory for the .ceos marker file. This file marks the root of the CEOS repository. If the project's CLAUDE.md specifies a CEOS/EOS root path, use that instead.

If .ceos is not found, stop and tell the user: "Not in a CEOS repository. Clone your CEOS repo and run setup.sh first."

Key Files (all relative to CEOS root)

| File | Purpose | |------|---------| | data/scorecard/metrics.md | Metric definitions (names, owners, goals, thresholds) | | data/scorecard/weeks/YYYY-WNN.md | Weekly scorecard entries | | data/meetings/l10/ | L10 meeting notes (primary data source) |

Process

Step 1: Determine the Week

Calculate the current ISO week (YYYY-WNN). Check if data/scorecard/weeks/YYYY-WNN.md already exists. If so, tell the user and ask if they want to update it.

Determine the week's date range (Monday through Sunday) for filtering queries.

Step 2: Read Metric Definitions

Read data/scorecard/metrics.md to get the list of metrics, their goals, and green/red thresholds.

Step 3: Synthesize from L10 Meetings (PRIMARY SOURCE)

This is the most important data source. Read all L10 meeting files from data/meetings/l10/ that fall within the scorecard week's date range.

For each L10 file in the week:

1. Read the full file
2. Extract from Headlines section: deal closings, revenue mentions, pipeline movement, content published, milestones hit
3. Extract from Scorecard Review section: any metric values already discussed
4. Extract from IDS section: context about what's working/not working
5. Extract from To-Do Review section: completed actions that map to metrics
6. Extract from Conclude / New To-Dos section: commitments that indicate activity

Build a synthesis per metric by scanning all L10s for the week. Match L10 content to each metric based on the metric's name and description from metrics.md.

Step 4: Auto-Pull from Available MCP Sources (optional, run in parallel)

Discover available MCP tools at runtime. Check which MCP servers are connected and query them for supplementary data. This step is entirely optional — the skill works with L10 data alone.

Common MCP sources and what to query:

| Source Type | Example Tools | What to Query | |-------------|---------------|---------------| | CRM (Attio, HubSpot, etc.) | search_records, search_records_advanced | Deal stage changes, pipeline value, revenue | | Calendar (Google, Outlook) | list_events, gcal_list_events | Meetings booked, discovery calls | | Email (Gmail, etc.) | search_messages | Correspondence related to deals | | Filesystem | Glob, Grep | Published content files, shipped deliverables |

If a CLAUDE.md specifies paths for content, episodes, or other deliverables, scan those paths for files with date prefixes matching the scorecard week.

If no MCP sources are available, skip this step entirely. The L10 synthesis from Step 3 is sufficient.

Step 5: Merge Sources and Resolve Conflicts

For each metric, merge the L10 synthesis with any external source data:

• L10 says X, external source says Y -- Present both, flag the discrepancy, let the user decide.
• L10 says X, no external data -- Use L10 value as the suggestion.
• No L10 data, external source says Y -- Use external value as the suggestion.
• Neither source has data -- Suggest 0, note "no data found -- manual entry needed."

Priority order: User's words in L10s > CRM data > Calendar data > Filesystem > Default to 0.

Step 6: Walk Through Each Metric (Sequential Questions)

This is the core interaction. Present ONE metric at a time using AskUserQuestion. Each question shows:

• The metric name and goal
• The synthesized suggestion with evidence
• The source(s) it came from
• Status based on the suggested value

Walk through all metrics in the order defined in metrics.md. For each metric:

1. Show a brief summary of what was found:

   ### [Metric Name] (Goal: [goal]/week)
   
   Suggested: [value]
   Sources:
   - [Day] L10: "[relevant quote or data point]"
   - [External source]: "[relevant data]"
   
   Status: [on_track/off_track] (goal: [goal], red threshold: [threshold])
   

2. Ask using AskUserQuestion with the suggested value as the default option:

   • Option 1: "[value] (suggested)" -- accept the synthesized number
   • Option 2: "Different number" -- user provides the correct number
   • Option 3 (if applicable): "0 -- didn't track" -- explicit zero

3. Record the confirmed value before moving to the next metric.

Key principle: The pre-populated suggestion should save the user time. If the L10s captured good data, the user just confirms. If not, they override. Either way, they only deal with one metric at a time.

Step 7: Compile Notes

After all metrics are confirmed, auto-generate the Notes section:

For each metric, create a note line:

• **[Metric]: [Value] ([status])** -- [source/context from L10s and external data]

Include a "L10 Context" subsection summarizing key themes from the week's L10s (IDS issues tackled, major headlines, patterns).

If external sources were used, include a "Data Sources" line listing what was queried.

Step 8: Show Final Scorecard and Write

Present the complete scorecard table:

| Metric | Owner | Goal | Actual | Status |
|--------|-------|------|--------|--------|
| ... | ... | ... | ... | ... |

Ask: "Save this week's scorecard?"

On confirmation, write to data/scorecard/weeks/YYYY-WNN.md using the standard CEOS template format:

---
week: "YYYY-WNN"
date: "YYYY-MM-DD"
---

# Scorecard -- YYYY-WNN

*Logged: YYYY-MM-DD (date range)*

| Metric | Owner | Goal | Actual | Status |
|--------|-------|------|--------|--------|
| ... |

## Notes

- **Metric: Value (status)** -- context
...

## L10 Context (synthesized from meetings)

Key themes this week:
- [theme from IDS discussions]
- [pattern across multiple L10s]

## Data Sources

*Synthesized from: L10 meetings ([N] this week)[, CRM (deals/stages)][, Calendar (meetings)][, filesystem (content/deliverables)].*

Step 9: Flag Off-Track Items

For any off_track metric, note: "Off-track items should be discussed during your next L10."

If 3+ metrics are off track, add: "Consider whether this week was an anomaly or a pattern worth investigating."

Offer to commit the scorecard file.

Guardrails

• Never write the scorecard without user confirmation. The whole point is interactive review.
• One metric at a time. Don't dump all metrics in one question. Walk through them sequentially so the user can focus.
• Show sources for every suggestion. Don't just show a number -- show where it came from (which L10, which data source).
• L10s are the primary source. They capture what the user actually said happened. External sources are validation/supplementation.
• Graceful degradation. If no L10s exist for the week, fall back to external sources only. If no MCP sources are available, ask for manual entry. Don't fail the whole flow.
• All paths relative to CEOS root. Never hardcode absolute paths. Use data/scorecard/weeks/, data/meetings/l10/, etc.
• Respect metric definitions. Use thresholds from metrics.md for status calculation.
• Keep it conversational. This runs in an interactive session. Talk to the user, don't just dump data.

Integration

• Use instead of ceos-scorecard when you want auto-populated suggestions for the "Log Weekly" mode
• ceos-scorecard is still available for "Define Metrics" and "Trend Analysis" modes
• After writing, suggest running this week's L10 if one hasn't been done yet