How We Work
Use this skill as an index. Read only the linked skill needed for the current action;
do not load every workflow by default.
Work tracking
Working modes
Companion skills from Matt Pocock's library, not org
skills. Install them separately (see the org-skills README). When they are installed as the
Claude Code plugin, prefix each name with mattpocock-skills:.
- Pressure-testing a plan, design, or scope before building: run
/grill-with-docs inside a
repository, or /grill-me when there is no working directory. The docs variant writes each
agreed term to CONTEXT.md and each hard-to-reverse decision to an ADR, so the next session
reads the vocabulary instead of re-deriving it.
- Planning an effort too large for one session, where the route to the goal is not yet visible:
run
/wayfinder. It charts decision tickets on the issue tracker and resolves one per
session. It produces decisions, not code, so hand off to the delivery skills below once the
route is clear. It reads docs/agents/issue-tracker.md, so run
/setup-matt-pocock-skills once per repository first and point it at Linear; without that
file wayfinder stops and asks for it.
Reach for a working mode before the delivery skills, not instead of them.
Delivery
Client communication
- Drafting a client message, status update, blocker, or recommendation: read
client-communication.
- Writing a client changelog entry, release note, or short fix summary: read
client-changelog.
Skill maintenance
If one task spans more than one category, read each relevant skill when that part of the
work starts.
Shared tool boundary
Use Linear MCP tools directly for Linear work and authenticated gh directly for GitHub
work. Use the configured authenticated GitLab tooling for a GitLab repository. Use Context7
for current library and framework documentation. Do not route these operations through a
generic wrapper CLI or task proxy. Existing account, team, repository, and task authority
remain the limit.
Board is truth · Act on reversible work · Surface blockers fast · Verify before review ·
Communicate plainly.
1---2name: how-we-work3description: Routes Diligence AI work to focused org skills for Linear workflow, task execution, pull-request delivery, client communication, changelogs, org-skill authoring, or the grilling and wayfinder working modes. Use at the start of a task, when blocked, or when unsure which team workflow applies. This is the lightweight index, not each full procedure.4---56# How We Work78Use this skill as an index. Read only the linked skill needed for the current action;9do not load every workflow by default.1011## Work tracking1213- Starting, resuming, updating, blocking, or completing work tied to Linear: read14 [linear-workflow](../linear-workflow/SKILL.md).15- Creating a new Linear issue: read16 [create-linear-task](../create-linear-task/SKILL.md).1718## Working modes1920Companion skills from [Matt Pocock's library](https://github.com/mattpocock/skills), not org21skills. Install them separately (see the org-skills README). When they are installed as the22Claude Code plugin, prefix each name with `mattpocock-skills:`.2324- Pressure-testing a plan, design, or scope before building: run `/grill-with-docs` inside a25 repository, or `/grill-me` when there is no working directory. The docs variant writes each26 agreed term to `CONTEXT.md` and each hard-to-reverse decision to an ADR, so the next session27 reads the vocabulary instead of re-deriving it.28- Planning an effort too large for one session, where the route to the goal is not yet visible:29 run `/wayfinder`. It charts decision tickets on the issue tracker and resolves one per30 session. It produces decisions, not code, so hand off to the delivery skills below once the31 route is clear. It reads `docs/agents/issue-tracker.md`, so run32 `/setup-matt-pocock-skills` once per repository first and point it at Linear; without that33 file wayfinder stops and asks for it.3435Reach for a working mode before the delivery skills, not instead of them.3637## Delivery3839- Deciding whether to plan, build, ask for a decision, or move to other work: read40 [task-execution](../task-execution/SKILL.md).41- Creating a branch or PR, reviewing, testing, marking ready, or merging: read42 [pull-request-workflow](../pull-request-workflow/SKILL.md).43- Running an independent Claude review: read44 [external-code-review](../external-code-review/SKILL.md).45- Verifying a completed feature in staging: read46 [staging-browser-qa](../staging-browser-qa/SKILL.md).4748## Client communication4950- Drafting a client message, status update, blocker, or recommendation: read51 [client-communication](../client-communication/SKILL.md).52- Writing a client changelog entry, release note, or short fix summary: read53 [client-changelog](../client-changelog/SKILL.md).5455## Skill maintenance5657- Creating or updating a Diligence org skill: read58 [org-skill-authoring](../org-skill-authoring/SKILL.md).59- Authoring, reviewing, or debugging any `SKILL.md`: read60 [writing-skills](../writing-skills/SKILL.md).6162If one task spans more than one category, read each relevant skill when that part of the63work starts.6465## Shared tool boundary6667Use Linear MCP tools directly for Linear work and authenticated `gh` directly for GitHub68work. Use the configured authenticated GitLab tooling for a GitLab repository. Use Context769for current library and framework documentation. Do not route these operations through a70generic wrapper CLI or task proxy. Existing account, team, repository, and task authority71remain the limit.7273> **Board is truth · Act on reversible work · Surface blockers fast · Verify before review ·74> Communicate plainly.**
Diligence-AI