Reference

Item types

An item is a bounded piece of work, a requirement, or a software capability. Use the smallest unit whose necessity can be judged on its own.

Type	Unit	Avoid
process	Work from a trigger to an outcome	Assessing a whole department as one process
step	An action within a process that changes something	One item per sentence in a written procedure
approval	Permission that work needs before it can proceed	Reading few rejections as no effect
meeting	A recurring arrangement where people meet	Assessing an attendee’s worth
report	An information product and the work of producing it	Requiring every record to trigger a decision
policy	A rule or obligation that people must follow	Equating a legal objective with its local implementation
software	A system that serves a purpose for its users	Equating the number of source files with unnecessary work
feature	A bounded capability a user can rely on	Treating every internal function as a capability
workaround	Work that compensates for a failure elsewhere	Removing the compensation while the failure remains

A file, function, screen, document, or interview excerpt is evidence about an item, not an item.

Kind	Means	Record
Documented	A source records a claim, procedure, or design	Where to find it and a quote
Reported	A person describes their experience or interpretation	Whether they saw it first-hand, and when
Observed	Records show behavior or outcomes	The period, the sample, and what it cannot show

A document shows what should happen. Confirm what does happen with a real case or someone who does the work.

Outcome	Requires to confirm
Keep	An owner
Change	An owner
Investigate	An owner
Stop	An owner, a scope, a review date, restart conditions, and no unresolved dependencies

Path	Contains
`skills/process-discovery/SKILL.md`	The runbook for a conversation
`skills/process-discovery/references/`	Method, scope, nine type modules, sixteen follow-up modules, decision and output templates, interviewing, saving
`skills/codebase-discovery/SKILL.md`	The runbook for code and data
`commands/`	`discover`, `map-codebase` and `conclude`
`agents/`	Seven agent roles for Claude Code
`.codex/agents/`	The same roles for Codex
`examples/purchase-approval.md`	An example conversation

Download the kit: discovery-kit.zip. Browse the method: method.md.

Role	Job
evidence-reader	Reads documents, logs, calendars, or code and returns labelled evidence. Read-only.
case-for-keeping	Builds the strongest case for keeping the work before a stop or change draft.
method-reviewer	Checks a summary against the limits below.
recorder	Saves the conversation to a connected workspace as one validated batch.
surface-surveyor	Lists the elements of one area of an application with the labels users see. Read-only.
code-tracer	Traces one element forward to what reads its value and backward to where it comes from. Read-only.
usage-measurer	Drafts read-only queries that measure use and turns the returned results into evidence. It never runs them.

Address: https://doweneedthis.com/mcp. Sign in with OAuth. Workspace membership is checked on every call. See Connect your AI tool.

Tool	Type	Use
`get_method`	Read	Returns the method index, or one module when you pass `module`
`list_workspaces`	Read	Lists the workspaces your account can access
`list_projects`	Read	Lists the projects in a workspace
`create_project`	Write	Creates an empty project
`read_map`	Read	Returns items, relationships, evidence, questions, answers, stakeholders, elements, flow nodes, and revision
`apply_discovery_batch`	Write	Saves items, relationships, evidence, questions, answers, stakeholders, software elements, and flow nodes against a revision
`read_decisions`	Read	Returns proposals, confirmed decisions, and follow-ups
`propose_decision`	Write	Creates a keep, stop, change, or investigate proposal
`list_baselines`	Read	Lists the frozen baselines, what was not final, and how far the map has changed since
`read_baseline`	Read	Returns a baseline’s findings as Markdown, or its raw map, decisions, or sessions
`read_outcomes`	Read	Returns each deliverable’s revisions, files, approvals, and the project’s progress
`register_outcome`	Write	Records a deliverable revision: its baseline and each file’s path, size, and SHA-256
`read_workshops`	Read	Returns session agendas, aggregate results, and feedback forms
`read_feedback`	Read	Returns unattributed written responses for one form

There is no tool to confirm a decision, freeze a baseline, approve a deliverable, change a subscription, or delete items.

Call read_map before every edit and send the revision with the save.
A save against an old revision is rejected. Read again and reconcile.
Repeat a save with the same requestId to retry it without creating duplicates. Use a new one for changed content.
Attribution is set by the server.
Stakeholders are labelled by role, never by name. Send the whole stakeholder record each time.
Flow nodes are the gateways, starts, and ends of a process. Join them to items with follows relationships and put each branch condition in the description of the relationship leaving the gateway.

Address	Purpose
`/app/`	The workspace
`/present/`	Presenter view for a live workshop
`/vote/`	A contributor’s personal voting link
`/feedback/`	A contributor’s personal feedback link
`/evaluate/`	A short solo diagnostic
`/mcp`	The MCP server

These apply to every conversation, record, and output.

Assess work, never people. No worth, no ranking, no “who is dispensable”.
No necessity scores, percentages, or grades. Counts of real things are fine.
No quota for finding waste and no push towards stopping.
Only people confirm decisions and set owners and dates.
Keep a workaround until its cause is fixed.
“Boring, manual, old, rarely used, unpopular, could be automated” are reasons to ask, not findings.
Votes show participants’ views. They are not evidence and not a verdict.
Feedback and workshop results show no names or identifiers.