GentBajko/capstone / skills/core
core
VerifiedInternal to the capstone suite - carries the shared rules (references/) and scripts every capstone subcommand reads; it exists so npx-skills installs ship them alongside the command skills. Not meant to be invoked directly; when invoked anyway, run references/../scripts/help.sh and output its stdout verbatim.
scripts/help.sh
View on GitHub#!/usr/bin/env bash
# Prints capstone usage. Single source of truth for the help text.
# Bash only: no PowerShell twin (lint-sync.sh check 8 enforces that).
cat <<'EOF'
capstone - codebase architecture reference generator
Usage: /capstone:<command> [subcommand]
Ten commands. Everything else is a subcommand: the seven pipeline
stages, reached as "start <stage>" when you want to enter mid-chain,
and the session review, reached as "review retro".
start Run the greenfield pipeline stage by stage:
mockup -> logic -> uiux -> architecture -> standards -> stack -> build
(resumes at the first incomplete stage; also
triggers on a bare "capstone" prompt)
start mockup Product discovery: seed + adaptive interview -> traceable markdown mockup
start logic Business-logic interview, scenario by scenario -> docs/capstone/logic/
start uiux How the UI looks and the UX behaves, from the mockup + logic
-> docs/capstone/uiux/
start architecture Greenfield: exhaustive architecture interview -> prescriptive reference
start standards Coding-standards interview -> normative standards.md
start stack Research libraries/services per capability, you pick -> 05-dependencies.md
start stack refresh
Re-vet the recorded picks: maintenance, license, pricing deltas
start build Implementation plan (backend then frontend), gate, then working code
map Build the reference docs, or refresh what drifted: no
reference yet builds one, an existing one gets only its
stale files rewritten, plus the logic/ scenarios and
uiux/ surfaces the map is missing
map rebuild Force a full rewrite of a reference that looks current
map <topic> Rebuild one chapter (architecture, models, conventions,
data-flow, dependencies, testing, operations, glossary,
interfaces)
map check Read-only trust report in two halves: a bash script
(staleness, ledger fragments, schema: stamps, headings
and their order, table columns, edge sites, payload
sections, model references, secret shapes ->
"MAP CHECK:") and the model's review (pointer drift,
absorption, re-vetting, coverage -> "MAP REVIEW:")
edges [folder] From the folder that holds several mapped repos: fill
each consumer's empty "from" in 09-interfaces.md out
of the code and the sibling clones. Asks nothing,
never writes a producer's "to", changes no answer
doctor [fix] Verify and repair the docs area: torn writes, index
drift, voided approvals, absorption gaps; "fix"
pre-approves the documented crash repairs
review [be|fe] Opt-in judgment -> review.md; no arg does both sides,
backend (architecture) and frontend (UI vs your design docs)
review retro [session]
The third judgment axis, and not a side: read a finished
session for evidence, then propose edits to standards.md and
your AGENTS.md/CLAUDE.md, one approved row at a time. Writes
no review.md and never touches code
groom <feature> Feature interview against the existing docs -> features/<date>-<slug>/spec.md
plan <feature> Task-by-task TDD plan from a groomed spec; gated on your approval
implement <feature> Execute the approved plan into code, in subagents or inline (you pick)
feature <desc> Take one feature idea to shipped code: chains groom -> plan ->
implement, resuming at the first unfinished stage
help Show this message
start and feature always ask inline or subagents on each new or resumed
run, before stage work. Inline avoids extra agent usage; subagents can
consume your allowance faster. No default: the run waits for your answer.
The choice covers every stage, review and reference refresh. A stage
entered directly as "start <stage>" asks too; pipeline handoffs inherit
the current choice.
Output: everything lands in docs/capstone/ - 00-index.md, the numbered
topic chapters, and the logic/ scenario map. Chapters carry the commit
they were derived at; the index carries none, so there is one copy of
freshness and `map` refreshes only what drifted.
Docs are strictly descriptive; only review judges.
Every command that writes records itself in docs/capstone/changelog.md.
A command that needs the reference and finds none builds it first.
CI gate: skills/core/scripts/map-check.sh [docs_dir] runs the script
half alone, no API key, checking every page against the output schema
in skills/core/references/schema.txt; templates/ carries both workflows.
Interview stages accept an optional artifact argument (a PRD, notes,
screenshots) that pre-fills answers for your confirmation.
Config: capstone.json in the agent's global folder (host-aware; ~/.claude
for Claude Code and ~/.codex for Codex, created at install): expertise 1-5 (vibe coder ... architect; how technical
conversations are, asked once then saved), teaching_mode (true =
narrate and teach while working), batch_questions (auto = an
interview's independent questions arrive together in one dialog, off =
always one per turn), docs_dir, index_file,
subagent_threshold, docs_in_git, language, redact (env-var name
patterns whose values the docs never quote), cross_repo (auto = map,
groom, plan and architecture consult quarry when it is on PATH).
The optional docs/capstone/capstone.json is the project's own shared
config, committed like the ledger: pipeline, workspaces (each name
doubles as the workspace's quarry target), or any global key to
override for that repo. expertise and teaching_mode are personal and
stay in the global file.
Cross-repo edges: 09-interfaces.md's frontmatter edges block is the
record. map writes kind, name, site and schema from the code and never
to/from - no repo holds another repo's name - quarry fills those by
joining the two halves on (kind, name), and you are asked only where
that join finds several candidates. With every repo cloned into one
folder, edges fills each consumer's empty from out of the code.
EOF
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 470k | Updated | Claude CodeCodex | Unverified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |