网有NETHAVE
AGENT SKILL

ctxt

Work with the ctxt CLI — an agentic context brain for capture, search, and composition on top of dPKMS (decentralized personal knowledge substrate). Automatically applies when: task involves capturing content, searching knowledge, composing briefs/plans, or managing focus profiles via the ctxt CLI. Use when: capturing content to knowledge base, running semantic search, generating compositions, managing config/profiles, or operating the dPKMS background worker. Description checklist: - [x] What it does: context-as-a-service CLI for humans and AI agents - [x] When to auto-activate: any ctxt/dpkms CLI operation - [x] Key surfaces: capture, search, compose, config, registry, entity - [x] Local-first, offline-capable, scriptable

使用场景Skill与智能体开发
License 许可
AGPL-3.0
最近更新
2026-11-09
Stars
★ 0
Forks
0

01 前言 Front Matter

name
ctxt
description
Work with the ctxt CLI — an agentic context brain for capture, search, and composition on top of dPKMS (decentralized personal knowledge substrate). Automatically applies when: task involves capturing content, searching knowledge, composing briefs/plans, or managing focus profiles via the ctxt CLI. Use when: capturing content to knowledge base, running semantic search, generating compositions, managing config/profiles, or operating the dPKMS background worker. Description checklist: - [x] What it does: context-as-a-service CLI for humans and AI agents - [x] When to auto-activate: any ctxt/dpkms CLI operation - [x] Key surfaces: capture, search, compose, config, registry, entity - [x] Local-first, offline-capable, scriptable

02 资源文件 500

03 SKILL.md 正文

ctxt CLI Skill

Teach an LLM agent to operate ctxt and dpkms effectively. Follow exactly; do not improvise flags or subcommand names.

Level 1 — Essentials (start here)

What is ctxt?

Two binaries, one system:

Binary Role Key concern
ctxt User-facing brain capture, search, compose, profiles
dpkms Infrastructure substrate server, jobs, storage, graph

All ingestion is async: ctxt enqueues a job → dpkms serve runs the pipeline → knowledge object is stored → available for search/compose.

Install

brew install contexthelp
# or
curl -sL https://context.help/install.sh | bash

Start the worker (required before any pipeline runs)

dpkms serve

Capture content (most common operation)

# Bare invocation → reads clipboard
ctxt

# Inline text
ctxt "The API rate limit is 1000 req/min" --hints "#api #limits"

# Piped stdin
echo "Fix signup flow" | ctxt --mention "@product.auth"

# From file
ctxt analyze --file notes.md --type text

# URL (auto-detected)
ctxt "https://example.com/blog/post"

Search

# Semantic (NLQ)
ctxt find "rate limiting strategies"

# No query → clipboard
ctxt find

# With profile lens
ctxt find "auth issues" --profile engineer

Compose

ctxt make brief --tag api --since 2026-01-01
ctxt make plan --mention "@product.auth"
ctxt make summary --profile founder
ctxt make draft --output report.md

Level 2 — Common Workflows

Workflow: capture → wait → inspect

# Capture and block until processed
ctxt "content here" --wait
# Returns knowledge object ID

# Open the result
ctxt open <id>

# Or list recent
ctxt list --limit 10 --sort recent

Workflow: triage inbox

# Capture to inbox (deferred processing)
ctxt analyze "rough idea" --inbox

# See what's pending
ctxt inbox list

# Promote to pipeline
ctxt inbox triage <id> --pipeline text.long

# Discard noise
ctxt inbox discard <id>

Workflow: query by filter

# By tag
ctxt list --tag api,security

# By mention
ctxt list --mention "@product.auth"

# By type + date window
ctxt list --type url --after 2026-01-01 --before 2026-03-01

# AST query
ctxt list --q "tag:api AND mention:@stripe.api"

# JSON output for scripting
ctxt list --json | jq '.[] | .id'

Workflow: entity graph exploration

ctxt entity list
ctxt entity show stripe.api
ctxt entity search "payment"
ctxt entity backlinks stripe.api    # what references this entity?

Workflow: registry management

ctxt registry list
ctxt registry add core https://registries.context.help/core
ctxt registry sync core
ctxt registry info core

Level 3 — Reference

Full command map

Command Purpose
ctxt [content] Capture (default; clipboard fallback)
ctxt analyze Explicit capture alias
ctxt inbox list/triage/discard/clear Inbox triage
ctxt job list/status/retry/cancel Job queue ops (over the API)
ctxt list Query knowledge objects
ctxt find <query> Semantic search
ctxt open <id> Show object detail
ctxt edit --id <id> Edit object metadata
ctxt delete Remove objects
ctxt make <type> Generate compositions
ctxt profile list/view/create/rm/set/unset Focus profiles
ctxt config show/path/validate/edit Config ops
ctxt registry list/add/remove/info/sync Registry ops
ctxt entity list/show/search/backlinks Entity ops
dpkms secret get/set/list Secrets backend
ctxt uri register Register ctxt:// OS handler
dpkms serve Start worker + REST/gRPC
dpkms housekeeping vacuum/reindex/compact/prune DB maintenance

analyze flags

Flag Description
--type <text|url|image|audio|video|feed|auto> Input type override
--hint "<#h1 #h2>" Influence tagging hints
--mention "<@slug1 @slug2>" Attach explicit mentions
--file <path> Read from file
--profile <name> Focus profile
--pipeline <name> Force pipeline
--raw Disable AI; store as-is
--wait Block until job completes
--inbox Defer to inbox
--output <json|yaml|id> Output format

list/find flags

Flag Description
--type <type> Filter by object type
--tag <t1,t2> Filter by tags
--hint <#h> Filter by hints
--mention <@slug> Filter by mention
--after / --before <ISO> Date window
--profile <name> Profile lens
--q "<query>" AST query
--limit <n> Max results
--sort <recent|match|score|weight|view> Sort order
--json / --yaml Machine output

make types

Type Output
brief Executive brief
plan Action plan
summary Summary
draft Publish-ready draft

Exit codes

Code Meaning
0 Success
1 General error
2 Invalid flags/input
3 Storage error
4 Registry error
5 Pipeline failure
6 Agent config error
7 Job execution error

Environment variables

Variable Purpose
CTXT_CONFIG Config path override
CTXT_DATA_DIR Data directory override
CTXT_PROFILE Default focus profile
DPKMS_DATA_DIR dPKMS data directory
DPKMS_WORKERS Worker thread count

Conventions

Mention syntax

Entities are referenced with @namespace.slug:

@product.auth     # entity "auth" in namespace "product"
@stripe.api       # entity "api" in namespace "stripe"

Use in --mention flags and --q AST queries.

Hint syntax

Hints are #tag-prefixed freeform influence signals (not formal tags):

--hints "#ux #mobile #p1"

Clipboard fallback

ctxt, ctxt analyze, ctxt find, and ctxt open all fall back to the system clipboard when no content/query/ID argument is given.

Two-phase ingestion

ctxt analyze returns a Job ID immediately. Use --wait to block and get the Knowledge Object ID. Or poll: ctxt job status <job-id>.

Config validation

Before running in CI or agent context:

ctxt config validate --check-secrets

Glossary

Term Meaning
Knowledge Object Core data unit; produced from any input by a pipeline
Mention @namespace.slug reference to a canonical entity
Entity Canonical concept with stable ID, aliases, relationships
Focus Profile Role/project lens affecting ranking, pipelines, surfacing
Pipeline Named sequence of steps transforming input → knowledge object
Registry Decentralized source of entity definitions / knowledge packs
Job Durable unit of async background work
Inbox Staging area for deferred-triage captures
Hint #tag-shaped signal influencing AI enrichment

Agent Patterns

Pattern: capture + extract ID for downstream use

id=$(ctxt "content" --wait --output id)
ctxt open "$id" --json | jq '.mentions'

Pattern: search → compose

# Find relevant objects, compose brief
ctxt find "auth rate limiting" --limit 20 --json > /tmp/results.json
ctxt make brief --tag auth --mention "@product.auth" --output /tmp/brief.md

Pattern: scripted batch capture

while IFS= read -r line; do
  ctxt "$line" --hints "#batch #import" --raw
done < items.txt

Pattern: agent worldview via profile

# All operations scoped to a profile
export CTXT_PROFILE=research
ctxt find "LLM memory approaches"
ctxt make summary --since 2026-02-01

Anti-Patterns

  • Never skip dpkms serve — pipelines do not run without the worker.
  • Never use --raw when enrichment is needed; it bypasses AI entirely.
  • Never assume synchronous processing; use --wait or ctxt job status.
  • Never pass secrets on CLI: dpkms secret set KEY takes the value from stdin, --from-file or --prompt, never an argument.
  • Never filter with both --tag and --q containing tag filters simultaneously (they compound; can produce unexpectedly narrow results).

04 同仓库更多 Skills

ContextHelp
ContextHelp/ContextHelp

tokenizer-step

Mock step

★ 0
ContextHelp
ContextHelp/ContextHelp

my-extractor

Mock step

★ 0
ContextHelp
ContextHelp/ContextHelp

ctxt-knowledge-retrieval

Train and prompt AI agents to use ctxt for deterministic knowledge retrieval with semantic fallback, provenance tracking, and reusable memory capture. Use when building agent loops over ctxt, debugging weak retrieval quality, or sharing insights, trial-and-error outcomes, and decisions across agents and sessions.

资料检索与研究Skill与智能体开发
★ 0
ContextHelp
ContextHelp/ContextHelp

pii-mask

Masks email addresses and phone numbers in captured text

邮件与沟通
★ 0