csk-adopt-plan-scaffold
Skill1 starsChanged 4 months agoMITView on GitHub
Exhaustive repository-local CSK adoption workflow for an existing codebase or project with implementation, plans, roadmaps, audits, backlogs, architecture, requirements, tests, handoffs, or product sources of truth. Use when .csk/project-state.json says adoption is pending or blocked, when a repository must be reverse-engineered before feature specification, or when existing evidence must become a provisional source-linked feature scaffold. Reads only the current repository and records completion or blockers through csk-start.
In this repository18 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- 1-csk-initSkill
- 2-csk-write-specSkill
- 3-csk-architectureSkill
- 4-csk-frontendSkill
- 5-csk-backendSkill
- 6-csk-qaSkill
- 7-csk-deploySkill
- audit-plan-loopSkill
- claude-skill-proxy-syncSkill
- create-worker-worktreeSkill
- csk-helpSkill
- csk-refineSkill
- csk-startSkill
- finish-branchSkill
- pull-main-ffSkill
- review-loopSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
20 rules. About 0 are common, 13 are distinctive, and 7 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈13
Distinctive · estimate
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 10
Treat the current repository as an existing product, not greenfield.
- Very specific: line 19
Run /csk-start state validation.
- Very specific: line 21
If state is pending first-use classification, return to /csk-start.
- Very specific: line 25
Read the canonical /1-csk-init skill and apply its adopt semantics when CSK product files are incomplete.
- Distinctive: line 34
Do not claim complete coverage until inventory, full-text reading, and mapping verification are complete.
- Distinctive: line 36
Do not substitute filenames, headings, search hits, or summaries for reading every relevant text artifact through EOF.
- Distinctive: line 38
Do not implement product code or create individual feature specs during the scaffold step.
- Very specific: line 47
Use git ls-files, rg --files, and targeted searches.
- Distinctive: line 72
Read every relevant text artifact through EOF in controlled batches.
- Distinctive: line 104
all relevant local indexes and artifacts were read.
- Distinctive: line 105
every discovered scope source maps to a feature candidate or an explicit non-feature classification.
- Very specific: line 111
If required local evidence is missing or unreadable, run the csk-start helper's state adoption-block --reason ..., add/update one durable …
- Distinctive: line 131
Do not create an empty greenfield list.
- Very specific: line 151
Every inventoried source appears exactly once.
- Distinctive: line 167
every relevant source maps to a candidate or explicit non-feature class.
- Distinctive: line 168
every relative link resolves inside the repository.
- Very specific: line 184
Ask one bounded question at a time and lead with a sourced recommendation.
- Distinctive: line 191
When the scaffold and its coverage are verified, run:
- Distinctive: line 207
After a confirmed candidate:
- Distinctive: line 216
Before exit, keep exactly one durable task for unfinished or blocked adoption work when repository writes are authorized.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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
CSK Adopt Plan Scaffold
Objective
Treat the current repository as an existing product, not greenfield. Inventory and read its relevant local code, plans, and sources of truth. Derive a provisional feature scaffold that preserves confirmed decisions, implemented maturity, known gaps, and source links. Review proposals with the user before returning to the numbered CSK workflow.
Entry contract
- Resolve the current Git root dynamically and read repository instructions.
- Run
/csk-startstate validation. Normal entry isproject/adopt/pendingorproject/adopt/blocked. - If state is pending first-use classification, return to
/csk-start; this skill must not make or store that decision implicitly. - Resolve source-of-truth paths only from repository-root
csk.config.jsonand.claude/rules/adapter-config.md. - Read the canonical
/1-csk-initskill and apply its adopt semantics when CSK product files are incomplete.
Boundaries
- Inspect only the current repository. Do not read another repository, global task store, message broker, or external workflow state.
- Preserve user changes, dirty worktrees, and files with unclear ownership.
- Change protected files only with concrete authority for that file.
- Do not claim complete coverage until inventory, full-text reading, and mapping verification are complete.
- Do not substitute filenames, headings, search hits, or summaries for reading every relevant text artifact through EOF.
- Do not implement product code or create individual feature specs during the scaffold step.
- Mark derived entries provisional. Technical implementation does not by itself prove product approval or release status.
- Preserve decisions already stored in repository files or made in the current conversation; reopen them only when contradictory evidence requires it.
Phase 1: Inventory repository evidence
Use git ls-files, rg --files, and targeted searches. Include tracked and
relevant untracked content; exclude .git, dependencies, caches, generated
output, builds, temporary files, vendored trees, and the starter-kit templates
themselves unless a project has intentionally filled them.
Search at least:
- implementation roots, package/module manifests, build files, and entry points;
- tests, fixtures, CI, release, packaging, and deployment configuration;
- plans, plan indexes, backlogs, portfolios, roadmaps, milestones, briefs, requirements, handoffs, and status records;
- PRDs, product vision, feature indexes, existing feature specs, user journeys, and acceptance criteria;
- architecture records, ADRs, diagrams, API or schema contracts, data-flow and security documentation;
- audits, reviews, remediation records, gaps, and operational runbooks;
- capability or gap inventories and other locally authoritative indexes.
For every candidate, record relative path, artifact class, title/purpose, discovery source, read status, and product relevance. Reconcile indexes with physical files: indexed-and-present, indexed-but-missing, present-but-unindexed, or supporting artifact rather than independent scope.
Phase 2: Read and classify evidence
Read every relevant text artifact through EOF in controlled batches. Inspect non-text artifacts with an applicable repository/runtime skill when their content is necessary. Extract:
- problem, objective, user or operational value, and product boundary;
- maturity and direct implementation/test evidence;
- dependencies, prerequisites, acceptance criteria, and non-goals;
- data, security, privacy, licensing, egress, and operating assumptions;
- open decisions, known gaps, conflicts, and supersession evidence;
- source links, commits, and other repository-local proof.
Classify each artifact as current, historical, superseded, resolved, cross-cutting, generated, vendored, tooling-only, or unresolved. Preserve contradictory claims with paths and evidence maturity. Prefer explicit ownership, supersession, and direct proof—not file dates alone.
Phase 3: Consolidate product knowledge
Separate evidence into:
- confirmed product decisions;
- derived current capabilities and maturity;
- open gaps, contradictions, and undecided contracts.
Define features by user outcome and testable product contract, not by renaming files or plan nodes. Several implementation artifacts may support one feature; cross-cutting evidence may support several.
Gate before writing
Write only when:
- all relevant local indexes and artifacts were read;
- every discovered scope source maps to a feature candidate or an explicit non-feature classification;
- known decisions and conflicts are recorded;
- existing feature files and the current Git diff were inspected;
- the adapter-resolved destination is confirmed.
If required local evidence is missing or unreadable, run the csk-start helper's
state adoption-block --reason ..., add/update one durable blocked task with the
exact recovery action, and stop. Do not present dependent claims as confirmed.
Phase 4: Write the provisional scaffold
Integrate with the adapter-resolved feature index, normally
features/INDEX.md; preserve useful existing content and unrelated changes.
Include:
- provisional-adoption status, date, source basis, and traceable counts;
- confirmed product basis and uncertainty labels;
- review groups by user journey or product boundary;
- unique provisional
PROJ-*IDs, titles, outcomes, priority, dependencies,Roadmapstatus, current evidence, and first known gap; - relative links from each feature to its supporting local sources;
- an explicit mapping for historical, resolved, cross-cutting, generated, vendored, and tooling-only sources;
- source-of-truth drift, cross-cutting gaps, review order, and next unused ID.
Do not create an empty greenfield list. Pre-fill what evidence proves and label everything else provisional.
Also write .csk/adoption-coverage.json as machine-readable proof:
{
"schema_version": 1,
"feature_index": "features/INDEX.md",
"sources": [
{
"path": "relative/source-path",
"feature_ids": ["PROJ-1"],
"non_feature_reason": null,
"manual_relevance_reason": null
}
]
}Every inventoried source appears exactly once. Map it to one or more real feature
rows, or use an empty feature_ids list plus a substantive
non_feature_reason in the form <class>: <specific justification>, where
<class> is one of cross-cutting, generated, historical, non-product,
resolved, superseded, tooling-only, or vendored and the justification
names why THIS source is not product scope (blanket one-word reasons are
rejected mechanically). A relevant source outside the deterministic inventory
also requires a substantive manual_relevance_reason; this makes the LLM's
semantic override explicit and reviewable. Never type summary counts by hand;
the deterministic tool derives them from this report and the current
inventory.
Phase 5: Verify mechanically and semantically
Verify:
- every relevant source maps to a candidate or explicit non-feature class;
- every relative link resolves inside the repository;
- feature IDs are unique and the next ID is correct;
- dependencies reference existing IDs and cycles are absent or explicitly open;
- status and maturity claims do not exceed evidence;
- historical or superseded material is not presented as current architecture;
- feature links and classifications semantically match the source content;
- the deterministic
state adoption-complete --coverage-report ...validation accepts the report, derives complete coverage, and rejects empty scaffolds; git diffcontains only intended changes plus preserved user work.
For a large inventory, run a machine-readable coverage check and report counts. A sample is not proof of complete coverage.
Phase 6: Review and persist the adoption result
Review feature candidates one at a time, starting with the earliest dependency. Ask one bounded question at a time and lead with a sourced recommendation. Cover user value, release scope, roles, data lifecycle, processing/egress, failure and abuse cases, quality/security/performance evidence, dependencies, and non-goals.
A candidate may be confirmed, renamed, split, merged, reprioritized, deferred, or rejected with a reason. Do not create every feature spec automatically.
When the scaffold and its coverage are verified, run:
csk_start.py state adoption-complete --coverage-report .csk/adoption-coverage.jsonThen reread .csk/project-state.json and the scaffold. Completion state must
point to the validated coverage report, its digest, the feature index, and
derived coverage counts. If completion cannot be proved, keep adoption pending
or blocked; never mark it complete optimistically.
Exit and continuation
Report inventory counts, scaffold path, candidate/review-group counts, coverage result, remaining drift or blockers, and the next candidate to review.
After a confirmed candidate:
/2-csk-write-speccreates its full spec./3-csk-architectureowns stack and design decisions./4-csk-frontendand/or/5-csk-backendimplement approved design./6-csk-qaverifies it./finish-branchsafely closes non-default branch work./7-csk-deployreleases only anApprovedfeature.
Before exit, keep exactly one durable task for unfinished or blocked adoption work when repository writes are authorized. Create no task for work expected to finish in the same turn.
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.

