Installs into .claude/skills of the current project.
Sponsors
SSkills DirectorySkills Directory
Know which skills are safe — weekly.
SSkills DirectorySkills Directory
Is that skill safe to install?
Are you the author of Creep Spread?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/brenpike-creep-spread)
---
name: creep-spread
description: Generates or enriches CONTEXT.md (or CONTEXT-MAP.md) with domain terms from project artifacts. Use to build a glossary.
allowed-tools:
- Read
- Write
- Bash(git rev-parse *)
- Bash(find *)
- Bash(ls *)
- Bash(wc *)
- Bash(jq *)
- Bash(test *)
- Bash(basename *)
- Bash(grep *)
shell: bash
---
# Creep Spread
Analyze a project's existing artifacts and generate or enrich a populated domain glossary in the canonical format of `${CLAUDE_PLUGIN_ROOT}/skills/plan-interrogation/references/CONTEXT-FORMAT.md`: for a new project adopting the hivemind plugin (seed-hive calls this skill), a project without a CONTEXT.md, or one whose CONTEXT.md needs newly discovered terms, such as after new modules or renamed concepts.
## Required Inputs
None. Operates on the project root from `git rev-parse --show-toplevel`, and needs at least one readable project artifact (README.md, CLAUDE.md, a package manifest, or source files).
## Procedure
1. Resolve the project root via `git rev-parse --show-toplevel`. Stop blocked if not a git repository.
2. **Mode detection** at the project root. Each update mode reads the existing files and puts their term names (lines matching the `**Term**:` pattern) into a known-terms set.
- `<project root>/CONTEXT-MAP.md` exists: `mode: update-multi`. Read it and every per-context CONTEXT.md it references.
- Else `<project root>/CONTEXT.md` exists: `mode: update`. Read it.
- Neither exists: `mode: create`. The known-terms set is empty.
3. **Resolve project name:** try in order:
a. `jq -r '.name // empty' package.json 2>/dev/null`
b. First `# ` heading in `README.md`
c. `basename` of project root directory, with hyphens/underscores replaced by spaces, title-cased
4. **Multi-context detection.** The directory heuristic scans `src/`, `services/`, `packages/`, `apps/`, `modules/`, `libs/`, `crates/`, `workspaces/` for directories containing 2+ subdirectories with source files; detecting 2+ bounded context directories is a hit.
- `mode: create`: `multi_context: true` on a hit, else `false`.
- `mode: update`: the root CONTEXT.md is authoritative for the single-context case per CONTEXT-FORMAT.md ("If only a root CONTEXT.md exists, single context"), so `multi_context` starts `false`. Still run the heuristic; a hit overrides it to `true`, so step 5 blocks for manual conversion.
- `mode: update-multi`: `multi_context: true` unconditionally. CONTEXT-MAP.md is authoritative for its existing entries: do not overwrite or re-detect them. Run the heuristic and flag each found directory not already listed in CONTEXT-MAP.md as a newly detected bounded context to add.
5. **Root CONTEXT.md conflict check:** if `multi_context: true` and `mode: update`, stop blocked with reason: "Root CONTEXT.md exists but project layout requires multi-context (CONTEXT-MAP.md + per-context CONTEXT.md files). Delete the root CONTEXT.md and re-run creep-spread, or proceed manually." Never auto-delete or auto-convert.
6. **Artifact discovery:** read each of these that exists, first 200 lines each:
a. `CLAUDE.md` — project rules, architecture, conventions
b. `AGENT.md` or `AGENTS.md` — agent instructions
c. `README.md` — project description, key concepts
d. `package.json`, `pyproject.toml`, `Cargo.toml`, `go.mod` — project name, deps
e. Top-level directory listing (`ls -1`)
f. Source file names in key directories (use `find . -maxdepth 3 -name '*.ts' -o -name '*.tsx' -o -name '*.py' -o -name '*.go' -o -name '*.rs' -o -name '*.java' -o -name '*.cs' -o -name '*.rb' | head -100`)
g. Grep for domain type declarations: `grep -rn 'class \|interface \|type \|enum \|struct \|trait \|model \|schema ' --include='*.ts' --include='*.py' --include='*.go' --include='*.rs' --include='*.java' --include='*.cs' --include='*.rb' -l | head -50` — capture filenames only, then read first 5 lines of each match for the declaration name
7. **Term extraction** from the step 6 artifacts, applying the Rules of `${CLAUDE_PLUGIN_ROOT}/skills/plan-interrogation/references/CONTEXT-FORMAT.md` (Rules):
- **Include:** Terms that represent concepts unique to this project's domain — entities, aggregates, value objects, domain events, bounded context names, business rules, workflows
- **Exclude:** General programming terms — controller, service, repository, handler, middleware, util, helper, config, error, response, request, manager, factory, provider, adapter, wrapper, base, abstract, interface (the keyword), type (the keyword), enum (the keyword), model (the keyword when used generically)
- **Group:** organize natural clusters (3+ related terms) under subheadings
- Give each term a one-sentence definition based on its use in the artifacts, and `_Avoid_:` aliases when synonyms are evident.
- **Deduplicate:** skip any term already in the known-terms set (case-insensitive), so re-running the skill never adds duplicates.
8. **Relationship extraction:** identify relationships between extracted terms, with cardinality where obvious (one-to-many, belongs-to, produces, consumes, etc.).
9. **Generate output file(s):**
**If `multi_context: false` and `mode: create`:**
Create `<project root>/CONTEXT.md` in the CONTEXT-FORMAT.md format:
```
# {Project Name}
{One sentence description derived from README or CLAUDE.md}
## Language
{Extracted terms with definitions and Avoid lists}
## Relationships
{Extracted relationships between terms}
## Example dialogue
> **Dev:** "{A realistic question using 2-3 of the extracted terms}"
> **Domain expert:** "{A clarifying answer that demonstrates term boundaries}"
## Flagged ambiguities
{Any terms that appeared with multiple meanings in the artifacts, or "None detected during bootstrap"}
```
**If `multi_context: false` and `mode: update`:**
Append new terms (not in the known-terms set) to the `## Language` section and new relationships to the `## Relationships` section of `<project root>/CONTEXT.md`. If no new terms were found, report `status: complete, terms_added: 0`.
**If `multi_context: true` and `mode: create`:**
Create `<project root>/CONTEXT-MAP.md` listing each bounded context with a relative path to its CONTEXT.md, and a `CONTEXT.md` inside each bounded context directory with that context's terms. Cap at 20 contexts.
**If `multi_context: true` and `mode: update-multi`:**
Add newly detected bounded contexts to CONTEXT-MAP.md. Append to each per-context CONTEXT.md the new terms not in that file's known-terms set.
10. Report output.
## Do Not
- Modify or remove any existing content in CONTEXT.md / CONTEXT-MAP.md; only append
- Remove terms
- Invent domain terms not evidenced in project artifacts
- Include general programming concepts as terms
- Read full source file contents; use structural signals (filenames, declarations, grep) only
- Commit, push, or touch git state
- Exceed 20 bounded contexts in multi-context mode
- Auto-convert a root CONTEXT.md to multi-context layout; stop blocked and inform the user
## Output
```text
status: complete | blocked
reason: <text> # when blocked
project_root:
- [absolute path]
project_name:
- [resolved name]
mode:
- [create | update | update-multi]
multi_context:
- [true | false]
files_created:
- [list of created files, empty when mode is update or update-multi]
terms_extracted:
- [count]
terms_added:
- [count of new terms added (0 when all discovered terms already exist)]
term_sources:
- [list of artifact files that contributed terms]
```
Use the schema from `${CLAUDE_PLUGIN_ROOT}/governance/report-format.md` (Worker Report — Blocked) for blocked states.