agentleFS
mrdaioffmrdaioff/baby-brain/.claude/skills/project-creation/SKILL.md

project-creation

Skill0 starsChanged 2 months agoMITView on GitHub

Use this skill when creating a new project folder under projects/, or when a conversation's context (a signed SOW, a new prospect, a recurring client relationship, a new internal build) implies one should exist but doesn't yet — even if the user hasn't explicitly asked to "create a project." Covers the three things every project needs to be discoverable (context.md, a registry.yml entry, a projects/README.md row), the files created only once they have something to hold (log, calls.md, open-items.md, README.md, a services/ catalog entry), the bucket-determination logic (the buckets in `_system/config.yml`), and the anti-patterns that leave a project half-created or padded with empty files.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Defaulting status to active: Only work actually under way is active.

    line 57
  2. Very specific:

    Starting calls.md blank when prior calls exist: Always search brand/artifacts/ and brand/transcripts/ first.

    line 58

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Project Creation Protocol

Creating a project is not just mkdir. A project is only "created" once it is discoverable two ways: by a human browsing projects/README.md, and by an agent matching new material against projects/registry.yml. Miss either and the next call or note about it gets filed somewhere else, or nowhere.

It is also not a form to fill in. Create the files that have something to hold. An empty call log or an empty stakeholders table is not a neutral placeholder: the next session reads it as "there are no calls" or "nobody else is involved", and may be wrong.

Before creating anything

  1. Determine the bucket. Which commercial identity is this work under? See buckets.list in _system/config.yml and the detail in .agents/work-contexts.md. Skip this if you work under a single banner, or if the project is not commercial work at all.

    Participant names alone are not enough. Your own partners and collaborators appear on calls across every bucket, so their presence proves nothing — it is the other side's names that identify an engagement. If genuinely ambiguous, ask; never auto-create into the wrong bucket.

  2. Pick the slug. Kebab-case. For work with a company, base it on the COMPANY name, not a contact's name — acme-logistics, not jane-doe: contacts change, the company is what the engagement is with. For your own work, name the thing — invoice-cli, pricing-research. Check projects/ for an existing folder or near-miss name first — don't create a duplicate under a slightly different slug.
  3. Search for prior material. Before writing anything, grep brand/artifacts/ and brand/transcripts/ for the names and terms involved. If prior calls exist, they must be reflected in calls.md and context.md from the start. A project that starts blank when months of real calls already sit in the repo is worse than no project — it reads as authoritative and is not.
  4. Determine actual status. active only if the work is actually under way — a signed SOW for client work, real commits for a build. Otherwise prospect. Don't default to active for optimism's sake — humans read the status field to decide what is real.

Always create

#FilePurposeTemplate source
1projects/<slug>/context.mdSnapshot a session reads on day one, under 400 lines: Current State (rewritten each update) / Key Decisions (one dated line each) / Open Questions / Notes & Nodes. No dated narrative — that goes in the log..agents/templates/project/context.md
2projects/registry.yml entrySlug, name, status, and the terms that identify this project's material: participants on the other side (if there is one) and project-specific keywords. This is how an agent filing new material recognises it as belonging here.The schema at the top of the same file
3projects/README.md rowOne row in the Active projects table, the index of every project.Same file

Create when there is something to put in it

FileCreate it whenTemplate source
projects/<slug>/log/YYYY-MM.mdThe first dated entry: a build-log entry, a call summary, the reasoning behind a decision. One file per month, append-only, entries headed ## YYYY-MM-DD — title..agents/templates/project/log.md
projects/<slug>/calls.mdThe first call. If step 3 found prior calls, that is now — seed them..agents/templates/project/calls.md
projects/<slug>/open-items.mdThe first commitment worth tracking — something someone owes, by a date. If step 3 found commitments in prior calls, seed them..agents/templates/project/open-items.md
projects/<slug>/README.mdOther people are involved, or the folder holds enough that someone opening it cold needs a map. Drop any section that would be empty..agents/templates/project/README.md
services/ catalog entryThe project has a SOW or scope. Check each deliverable against services/README.md — reinforce an existing entry or add one. See .agents/skills/services-catalog/SKILL.md.Same skill
(CLI-specific) private memory pointerThe CLI you are running has its own cross-session memory store. A cache, never the record — invisible to every other CLI..agents/memory/index.md

A software project, a research thread, or anything with no other party will often be just context.md and a log for a long time. That is the correct shape, not a half-created project.

When you do create one of these files, copy its template — do not improvise it. Keep every heading and column order exactly as the template has it. Two of them are read by scripts:

  • open-items.md — scripts/rollup-open-items.mjs reads the ## Open Items table and stops at ## Closed. Rename a heading or reorder a column and this project silently vanishes from projects/OPEN-ITEMS.md. No error — it just stops appearing.
  • context.md — the health scan flags any context.md over 400 lines, which is what happens when dated entries get stacked in it instead of in the log.

Everything a future session actually needs must be in the committed files, not in private memory. A fact that lives only in one CLI's memory store is lost the moment the owner opens a different tool. If something portable surfaces while creating the project, put it in context.md or the relevant .agents/ file.

Commit discipline

Stage only the files this task actually touched — never git add -A in this repo, given how much untracked scratch/export data typically sits in projects/*/. New commit, no --no-verify. For push behaviour and the concurrent-session verification step, follow .agents/working-agreements.md § Git and delivery — that file is the authority.

Anti-patterns

  • Creating context.md alone and stopping. Without the registry entry and the index row, nothing finds the project: a human browsing the index misses it, and the next call about it gets filed elsewhere.
  • Creating every template "to be complete". Four files where three are empty tables tell the next session there is nothing to record. Create a file when it has its first entry.
  • Defaulting status to active. Only work actually under way is active.
  • Starting calls.md blank when prior calls exist. Always search brand/artifacts/ and brand/transcripts/ first.
  • Duplicating context.md's decision log into a private memory file. Memory is a pointer + the handful of facts that change how you'd behave — not a second copy of the project history.
  • Skipping projects/registry.yml. The easiest one to forget, because nothing breaks immediately — the next call or note about this project just gets filed somewhere else, and someone has to notice and backfill it later.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.