bulma-memory
Skill1 starsChanged 49 days agoMITView on GitHub
Use when /memory, /bulma-memory, a bulma graph or engine starts, the first run in a repo, CODEOWNERS or AGENTS.md list no reviewers, or the user states a durable preference (chat language, reviewers, project notes, open questions).
In this repository35 other files it gives its agents.
- AGENTS.mdAGENTS.md
- skills/bulma-feature-delivery/adapters/claude.mdCLAUDE.md
- architectSkill
- before-and-afterSkill
- blast-radiusSkill
- bulma-busSkill
- bulma-code-reviewSkill
- bulma-developerSkill
- bulma-fdSkill
- bulma-feature-deliverySkill
- bulma-goalSkill
- bulma-hostSkill
- bulma-lstmSkill
- bulma-qaSkill
- bulma-reviewerSkill
- bulmaSkill
- bulma-triageSkill
- diagnoseSkill
- figure-it-outSkill
- grill-with-docsSkill
- howSkill
- interrogateSkill
- kevinSkill
- no-commentsSkill
- principle-fix-root-causesSkill
- principle-prove-it-worksSkill
- principle-sequence-verifiable-unitsSkill
- receiving-code-reviewSkill
- tddSkill
- technical-writingSkill
- thermo-nuclear-code-quality-reviewSkill
- to-issuesSkill
- to-prdSkill
- typescript-best-practicesSkill
- unslopSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 3 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 36
No git root? Only the home file.
- Very specific: line 41
Before the first user-facing sentence of any bulma graph, engine, or /memory:
- Distinctive: line 45
Read the home file, then the project file, then the worktree file when it is a different path.
- Very specific: line 48
If project ## Open has an item and this turn is not wait_ci / execute / a last-resort ticket ASK: ask one Open item at the end of the turn.
- Very specific: line 59
If absent, use $BULMA_HOME/workspaces, where BULMA_HOME defaults to $HOME/.bulma.
- Very specific: line 140
Do not duplicate a fact that already lives in the target repo (AGENTS.md, CODEOWNERS, REVIEW.md).
- Distinctive: line 175
Write memory inside a repo
- Very specific: line 199
When authorized by the user, project ## Preferences may also record feedback_issues: publish and feedback_repo: owner/repo.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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.
Files
Bulma memory
Durable prefs outside git. Not STATE. Not AGENTS.md.
Paths: DISK.md.
$BULMA_HOME/memory.md # you (every repo)
$BULMA_REPO/memory.md # this repository, shared by all its worktrees
$BULMA_ROOT/memory.md # this worktree only, when it differs from $BULMA_REPO$BULMA_REPO is the $BULMA_ROOT of the repository's main worktree, so a
task worktree (Orca, Codex, Grok, git worktree add) reads the same project
memory as the main clone:
main=$(git worktree list --porcelain | sed -n '1s/^worktree //p')
BULMA_REPO="$BULMA_HOME/$(printf '%s' "$main" | sed 's|^/||; s|/|-|g')"In the main clone $BULMA_REPO and $BULMA_ROOT are the same directory.
"Project file" below means $BULMA_REPO/memory.md.
No git root? Only the home file. Create a file when writing, not on install. Missing file = empty.
Load
Before the first user-facing sentence of any bulma graph, engine, or
/memory:
- Resolve
$BULMA_HOME,$BULMA_ROOT(DISK.md) and$BULMA_REPO(above). - Read the home file, then the project file, then the worktree file when it is a different path. Later files win on the same key; sections merge.
- Chat language = home
## Chatlanguage:. Defaulten. - If project
## Openhas an item and this turn is notwait_ci/execute/ a last-resort ticket ASK: ask one Open item at the end of the turn. Do not stop the graph for it.
Workers that do not chat skip this. Ship still resolves reviewers via PRODUCT.md §6.
Worktree location
The Git fallback worktree directory is a memory preference. Read
worktree_root from ## Preferences using the normal memory precedence.
If absent, use $BULMA_HOME/workspaces, where BULMA_HOME defaults to
$HOME/.bulma. Installation creates this default directory without creating
or rewriting memory files.
For a custom location, write to home memory, or project memory for a repo-only override:
## Preferences
- worktree_root: /absolute/path/to/workspacesResolve the value to an absolute path and pass it as WORKTREE_ROOT to
JOBS.md. A leading ~, $HOME or
$BULMA_HOME denotes that directory; expand it as path text, never with
eval or command substitution from the memory value. Create the resolved
root before Git fallback creation. A custom location does not move existing
checkouts or change paths chosen by host isolation or Orca.
Chat vs forge
| Surface | Language |
| Chat with the user | home ## Chat (default English) |
| GitHub / GitLab (PR body, review, thread, issue comment, dismiss) | English |
| Spec, tickets, commits, CI, skill bodies | English |
Apply Kevin for human-facing communication. A PT-BR chat pref does not translate forge text.
Files
Same headings. Write ## Chat only in the home file. Write
## Product and ## Open only in the project file.
# Memory
## Chat
language: pt-BR
## Preferences
- short durable notes
- verification_recipe: docs/verification.md
## Product
reviewers: alice, bob
## Gotchas
- src/billing/**: amounts are minor units; divide once, at display
- duration: stored in seconds everywhere
## Checks
- python3 tools/prcheck.py --base origin/main
## Open
- Confirm reviewers: alice, bob (merged PRs)language: is en or pt-BR (or another BCP-47 tag). Reviewers are
forge logins, comma-separated, never invented. ## Product is the
confirmed reviewer list. How it gets filled: PRODUCT.md §6.
second_opinion: under ## Preferences turns on a second critic on another
model for elevated-risk plans: codex, gemini, or claude. Leave it out
for off. It names a CLI installed on this machine, so it is usually a user
preference, not a project one. How it runs: bulma-host §Second opinion.
## Gotchas and ## Checks live only in the project file.
## Gotchas: one line per domain fact the code does not make obvious (units, id conventions, provider quirks, fields that look alike). Start each line with a path glob or a keyword, then:and the fact. The coder gets the lines whose glob or keyword matches its files; the reviewer gets the lines that match the diff. A gotcha is a lead, never a finding by itself. Add one when a review or an incident shows a fact that the next change will trip on again.## Checks: one command per line that this project runs before a push (a guardrail scanner, a schema validator). The fd tester runs them from the repo root after its own commands. Never put a deploy or a destructive command here.
Do not duplicate a fact that already lives in the target repo
(AGENTS.md, CODEOWNERS, REVIEW.md).
/memory
/memory
/memory me responder e interagir comigo apenas em PT-BR
/memory --project reviewers: alice, bob| Args | Write |
| empty | Print both files. Ask the first ## Open item if any |
--project / clearly this-repo (reviewers, assignee, repo notes, gotchas, checks) | project file |
| else | home file |
Parse the note. Language / idioma / PT-BR / Portuguese → home
## Chat. reviewers: logins → project ## Product (that is
confirmation). gotcha: → project ## Gotchas. check: → project
## Checks. Other text → ## Preferences on the chosen file.
Append or replace the matching key. Do not rewrite unrelated sections. Do not commit these files.
Open
Graphs may append one line under project ## Open when a fact is
useful later and lookup failed. One question per turn. After the
answer: write the fact, delete that Open line, continue.
Open is not grill. Ticket ASK stays DECISION_POLICY.md.
Never
- Write memory inside a repo
- Hardcode a company's handles in a graph
- Copy a project's gotchas or checks into a skill file
- Block CI, ship, or execute to drain Open
- Interview a first-run questionnaire
- Let chat language leak onto the forge
Verified prevention
## Prevention lives only in project memory. Load it with other project
sections. Each entry has a stable pattern key, stages development, review
and/or qa, path glob or flow keyword, concrete check and source evidence.
Pass only matching entries to the affected stage. Missing entries add no work.
Rules guide investigation and never replace current evidence or product ACs.
## Prevention
- save-path-parity | development,review,qa | editor/save | Check manual save and autosave reject the same invalid input. | source: <verified finding URL and SHA>LSTM and reviewer maintain these through FEEDBACK.md.
Reuse an existing rule instead of copying it here. Keep domain facts under
## Gotchas. Concurrent workers hand shared writes to the coordinator.
When authorized by the user, project ## Preferences may also record
feedback_issues: publish and feedback_repo: owner/repo. Without publication
authorization or a known destination, feedback tasks remain local drafts.
These settings never authorize issue comments, product fixes or skill rewrites.
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
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.

