Skip to content
Back to skills

Commitments

ASecurity

Use to track promises in ~/haku-work-reflections/commitments.md and keep them from going stale. Trigger phrases include "what have I committed to", "add a commitment", "I promised X by Friday", "am I overcommitted", "help me say no to this", "I need to renegotiate a deadline". Renegotiate early; pressure-test new asks. Distinct from wins-log (what you did vs. what you owe).

  • 7 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added September 22, 2026
toolsrustgoreactgit

Works with

  • cli

Security analysis

A100/100

Scanned September 22, 2026

npx -y skills add shrwnsan/vibekit-claude-plugins --skill commitments --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Commitments?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Commitments
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/shrwnsan-commitments/badge)](https://www.skillsdirectory.com/skills/shrwnsan-commitments)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: commitments
description: Use to track promises in ~/haku-work-reflections/commitments.md and keep them from going stale. Trigger phrases include "what have I committed to", "add a commitment", "I promised X by Friday", "am I overcommitted", "help me say no to this", "I need to renegotiate a deadline". Renegotiate early; pressure-test new asks. Distinct from wins-log (what you did vs. what you owe).
---

# Commitments

Trust upward is made of kept promises and *early renegotiation* of unkeepable ones. The failure mode isn't the missed deadline — it's the missed deadline the other person discovered on the deadline. This skill keeps a ledger of what the user owes and makes renegotiation happen while it's still cheap.

## The ledger

`~/haku-work-reflections/commitments.md` (respect `$HAKU_WORK_REFLECTIONS_HOME`). One entry per promise:

```markdown
## [C-014] Eval regression report to Priya (VP Eng)
- promised: 2026-07-08, in the quarterly review
- due: 2026-07-18
- status: open          # open | done | renegotiated | dropped
- stakeholder: priya-shah (managing-up)
- notes: she cares about the enterprise-tier number specifically
```

Statuses are never deleted, only updated — the history of renegotiations *is* the useful data.

## Mode 1: Log a promise

When the user says they committed to something (or `exec-readout-prep` / a 1:1 just ended with a promise made): capture what, to whom, by when, and where it was said. Thirty seconds. If the counterparty is a registered stakeholder, link the slug.

Push back once on vague promises at capture time: "look into it" has no due date and no done-state — *"what did they hear you promise?"* is the question. Log what they heard, not what you meant.

## Mode 2: Review ("what do I owe?")

On request, or as a glance inside `weekly`:

1. **Due soon** (next 7 days) and **overdue**, sorted by counterparty seniority and age.
2. **The slipping** — open items with no progress signal the user can name. These are renegotiation candidates *now*, while the due date is still in front of them.
3. **Load read** — total open commitments. If the count only ever grows, say so plainly: that's not a busy quarter, that's an accumulating default.

For each slipping item, one recommendation: deliver, renegotiate (see script), or consciously drop with notice. "Hope" is not on the list.

## Mode 3: Renegotiate early

The script that preserves trust:

> *"I committed to [X] by [date]. [One-sentence reason it moved — no essay, no blame theater.] I can do [scoped-down version] by [original date], or the full thing by [new date]. Which is more useful to you?"*

The two-option structure matters: it converts an apology into a decision the counterparty controls. Renegotiated early, this *builds* credibility. Update the ledger entry with the new terms and date.

## Mode 4: Before saying yes (or no)

When a new ask arrives, read the open ledger first and put the trade-off on the table:

> *"I can take that. Here's what's currently committed: [top 2–3 relevant items]. If this is urgent, which of these should move?"*

This is the honest no: not a refusal but a prioritization question routed to the person entitled to answer it. If the user simply wants to decline, help them decline once, clearly, with a reason and (when real) an alternative — not a slow-fade maybe. For asks that deserve a yes, log the commitment immediately (Mode 1) so the yes was priced.

## Anti-patterns

- **The ledger as guilt engine.** Overdue items get a renegotiation plan, not a shame spiral. The skill's tone is operational, always.
- **Tracking tasks.** This is not a todo list. Only promises *another person is relying on* go in; the bar is "would someone notice if this didn't happen?"
- **Renegotiating everything.** A user whose entries are majority-renegotiated has a calibration problem, not a communication problem. Surface it once, kindly: the fix is smaller promises, made later.
- **Secret drops.** Dropping a commitment is legitimate; dropping it without telling the counterparty is how trust dies. `status: dropped` requires "notified them" in the notes.

## Composition

- `weekly` — the natural place for the Mode 2 glance
- `exec-readout-prep` — approved asks come with promises; log them on the way out
- `one-on-one-prep` — commitments due to the person you're meeting belong on the agenda
- `the-status-crafter` — "next period" lines in status updates are commitments; keep the two consistent
- `stakeholder-reflect` — a counterparty's reaction to a renegotiation is worth a dated entry

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.