stark-ai-de/agent-skills / plugins/stark-ai-developer/skills/karakeep-knowledge
karakeep-knowledge
UnverifiedSet up or use a local persistent Karakeep knowledge collection through its official MCP server. Use when configuring Karakeep for coding agents, checking an existing connection, checking project readiness, saving factual session reports, synchronizing approved canonical documents and separate drafts, maintaining approved source-backed aggregates, curating lists and tags, or consulting saved web sources for development work.
references/code-records.md
View on GitHub# Bounded code-record grant
The optional `codeRecords` field extends schema-version 1 collection policies. Existing policies retain their behavior. `kinds` must explicitly include `code-record`, destination list IDs must be the adopted private lists, roots and registry file must be absolute, include/exclude patterns must be nonempty relative scopes, `maxItems` is 1–20, `maxSeconds` is 1–600, and `approval` records the actual user grant. See the example asset. Model agreement is not adoption.
Use the existing connection, credential file, account state root and writer host. The shared writer lock, ownership receipts, conflict handling, readback and uncertain-response recovery remain authoritative. No new network client or independent writer is introduced. Approved findings/dossiers keep their original aggregate-grant rules; a code record cannot be used as original aggregate evidence.
An intent item has `kind: "code-record"`, a stable component `id`, `title`, and `codeRecord` with `selection`, captured `source`, `problem`, `useCase`, `dependencies`, `license`, `tests`, `adaptation`, `status`, and ISO `reviewedAt`. Selection contains registered `repoId`, relative `file`, `symbol`, and one-based `startLine`/`endLine` (at most 501 lines). Import `captureCodeSource(policy, selection)` from `scripts/knowledge/code-records.mjs` to capture proof. This read is local. The planner and each mutation independently recapture it; drift requires a new plan.
Source evidence includes repository identity, exact file/symbol/range, file and excerpt SHA-256 fingerprints, and a revision only if the actual file bytes match that HEAD file. Untracked or changed files are local drafts. Missing files can retire only an already owned record; an invalidated line range cannot remain validated. File symlinks, oversized/non-UTF8 files, unregistered roots and excluded/secret paths fail closed.
Identity is policy + repository + component, not a commit. Status is `candidate`, `source-checked`, or `practically-validated`; rendering explicitly attributes assessment to an agent. Tests use `command` and `outcome` (`not-run`, `passed`, `failed`). Executed claims also require a private `evidenceFile` and its SHA-256 `evidenceHash`: JSON schemaVersion 1, matching command/repoId/sourceHash and integer exitCode. Practical validation needs a passed receipt. The helper verifies evidence but neither runs test commands nor certifies a human review.
The normal managed-section three-way merge preserves user text outside the owned section and blocks conflicting changes inside it. Recheck sources periodically, downgrade changed components pending semantic reassessment, and surface retired/unresolved sources. A provider timeout is uncertain, not permission to retry a create. Follow the existing recovery protocol and receipts. A busy lock is pending; never steal a live lock. The apply deadline is checked before every remote mutation and source is rehashed; an in-flight bounded HTTP request may finish after the deadline.
Provenance also captures the current branch, a credential/query/fragment-free
HTTP(S)/SSH/Git origin, and the nearest regular conventional license file (relative
path and SHA-256, at most 64 KiB). Local/unsupported remote schemes are omitted.
No license file means `unknown`; an agent's free-text license assessment is not a
license grant or verified legal classification. Inspect the referenced license
and its applicability before reuse. Detached HEAD is explicit.
`draft`, `needs-review`, `stale`, and `retired` are rendered conventions, not native
Karakeep statuses. `candidate` or uncommitted content is a draft, assessed current
content needs human review, invalidated assessment/ranges are stale, and removed
sources are retired. Optional `assessmentSourceHash` binds the prior technical
assessment: a mismatch requires candidate status and no inherited executed-test
claims. Human review and manual notes remain outside the managed section and are
preserved. Agent assessment never sets human approval. A rename retires the old
selection; an explicitly verified new selection can retain the same component ID.
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: 471k | Updated | Claude CodeCodex | Verified | 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 |