calle-ai/awesome-phone-call-agents/call-agent-commitment-tracker

call-agent-commitment-tracker

**An agent that says "I'll send that right away" and doesn't — is worse

than one that said nothing.**

When an automated agent makes a call, it often commits to future actions: sending a confirmation email, having a specialist call back, filing a referral. Those commitments are audible, remembered by the person, and completely untracked unless this skill reads the transcript.

This is the post-call layer: it reads the finished get_call_run transcript, extracts every agent commitment, classifies it by urgency, and hands you a ready-to-use follow-up call goal to verify fulfillment.

When To Use

• after any CALL-E call where the agent may have made forward-looking promises
• as part of a quality-assurance pipeline to prevent "ghost commitments"
• before a follow-up call to understand what the previous call committed to
• in collection, healthcare, or support workflows where broken commitments

carry legal or reputational risk

When Not To Use

• to audit the callee's promises (this skill only tracks agent turns)
• during a call; strictly post-call analysis plus pre-call goal crafting
• as a definitive legal record; it is heuristic and advisory only
• to replace CRM or ticketing systems for obligation tracking

Workflow

Audit a finished call

python3 scripts/commitment_tracker.py analyze \
  --transcript path/to/call-result.json

Reads the real get_call_run result shape ({status, result: {transcript}}) or the flat fixture shape used by sibling skills. Emits a commitment card:

• commitments[]: each detected commitment with turn_index, classification,

and evidence (PII-masked, capped at 200 chars)

• commitment_count: counts by WITH_DEADLINE, WITHOUT_DEADLINE, CONDITIONAL
• verdict: COMMITMENTS_FOUND / NONE_FOUND, plus unclear paths

(empty transcript, no agent turns)

• recommended_action: schedule_followup_call with guidance, or

no_followup_required

• disclaimer: heuristic advisory disclaimer on every card

Commitment classifications

Callee turns are always excluded

The skill only scans AGENT turns (speakers not in callee / customer / patient / caller / recipient / user). A callee saying "I will think about it" is never recorded as an organization commitment.

Craft the follow-up call goal

python3 scripts/commitment_tracker.py craft --scenario commitment-followup

Emits the plan_call inputs JSON whose goal instructs the next call to: verify whether the promised action was carried out, acknowledge any gap without blame, and escalate to a human colleague if needed — without making new commitments itself.

Scientific Foundation

This skill implements a lexical/regex heuristic that operationalizes the commissive-speech-act class. It does not use model internals and labels every output analysis_mode: "heuristic".

Differences from sibling skills

• call-cross-call-consistency-checker compares stated facts across two

calls for contradictions; this skill tracks future-action pledges within one call.

• call-review audits general call quality and compliance; this skill

specializes exclusively in forward-looking agent commitments.

• call-agent-certainty-calibrator grades how confidently the agent stated

facts; this skill grades whether the agent made promises it should follow up on.