---
name: obligation-extraction-tracker
description: Extract every obligation from signed contracts with two independent passes, then track deadlines automatically. Use this skill when: New signed contract, or weekly sweep — Added to the executed-contracts folder, or the Monday catch-all.
---

# Never miss an obligation in a signed contract

Goal: extract every obligation from a signed contract with two independent passes and track it to deadline.

Connected tools this skill uses: drive, googlesheets, slack, gmail.

## When to use

- New signed contract, or weekly sweep — Added to the executed-contracts folder, or the Monday catch-all.
- Or when the user asks directly to extract every obligation from a signed contract with two independent passes and track it to deadline.

## Method

Extract every obligation from {{contract_path}} and track it to deadline.

1. Parse {{contract_path}} with the document skill into full text.
2. Run two independent extraction sub-agents over the text — pass A and pass
   B — each identifying every obligation: what's owed, by whom, and by
   when. Neither pass sees the other's output.
3. Validation: reconcile pass A and pass B. Where both agree, accept the
   obligation. Where they disagree or only one pass found it, run a third,
   targeted re-read of that clause before deciding — never publish an
   obligation neither pass independently found.
4. Write each reconciled obligation to the obligation-extraction-tracker
   view: description, owner, due date, source clause, and a link to the
   contract.
5. Schedule a deadline reminder job for each obligation at
   {{reminder_lead_days}} days before it's due, and one on the day itself.
6. Post a summary to {{delivery_channel}}: obligations added, obligations
   sent to the third read, and anything that stayed unresolved for
   {{owner}} to check manually.

Never mark an obligation satisfied or close a reminder without a human
confirming it — the tracker flags, it does not certify compliance.

## Obligation extraction

Identify every deliverable, payment term, reporting cadence, and deadline a signed contract creates.

## Method
1. Read the parsed contract text from the document-parsing skill.
2. Scan for language that creates an obligation: deliverables, payments, reports, notices, and recurring cadences.
3. For each obligation, capture what is owed, who owes it, and the due date or trigger.
4. Distinguish an obligation from background language that describes rights without creating a duty.
5. Exclude renewal-window language, which is tracked separately by the renewal-radar recipe.
6. Record the source clause for every obligation extracted.

## Rules
- Only extract obligations the contract text actually creates, never an inferred best practice.
- Renewal and notice-window dates are out of scope here, to avoid double-tracking with the renewal radar.
- What counts as an obligation versus background language (set on first run, then stored in memory).
- Every obligation must carry a source clause; an obligation without one is not published.

## Output
A list of obligations, each with description, owed-by, due date or trigger, and source clause, ready for the reconciliation pass.

## Document parsing

Convert a signed contract in Word or PDF form into full text for obligation extraction.

## Method
1. Fetch the contract at {{contract_path}} from the executed-contracts folder.
2. Convert the file to full text, keeping clause and page references intact.
3. Handle both docx and PDF sources without losing section structure.
4. Flag any file that fails to parse cleanly rather than passing partial text downstream.
5. Hand the parsed text to both extraction sub-agents identically, so neither has an advantage.

## Rules
- Both extraction passes must receive the exact same parsed text, with no version drift between them.
- A parse failure is reported, not silently skipped from the tracker.
- Executed-contracts folder location (set on first run, then stored in memory).
- Preserve clause and page references so every published obligation can cite its source.

## Output
Full contract text with clause and page references, delivered identically to both extraction sub-agents.

## Reconciliation policy

Resolve disagreement between the two independent obligation-extraction passes before anything is published.

## Method
1. Compare pass A and pass B obligation lists for the same contract, matching by clause and description.
2. Where both passes independently found the same obligation, accept it.
3. Where only one pass found an obligation, do not accept or reject automatically — queue it for a third read.
4. Run a third, targeted re-read of the specific clause in question.
5. Decide from the third read whether the obligation is real; publish or discard accordingly.
6. Log every disagreement and its resolution for later review of extraction quality.

## Rules
- Never publish an obligation neither pass independently found.
- A single-pass finding always goes to a third read before a decision, never straight to publish or discard.
- Track disagreement patterns by clause type (set on first run, then stored in memory) to catch a systematically weak pass.
- The tracker records obligations as flagged, never as satisfied — only a human confirms completion.

## Output
A reconciled obligation list ready for the tracker view, plus a disagreement log noting what triggered each third read and its outcome.

## Recurring jobs

This skill pairs with scheduled jobs. Offer to set each one up; only create the ones the user approves:

- Executed contracts watch (Every 30 minutes on weekdays, check the executed-contracts folder) — https://eluu.ai/cookbook/obligation-extraction-tracker/jobs-setup/executed-contracts-watch.md
- Weekly obligation catch-all sweep (Every Monday morning) — https://eluu.ai/cookbook/obligation-extraction-tracker/jobs-setup/weekly-obligation-catch-all-sweep.md