Index of every page on this site: llms.txt
All skills
relatedb avatar

/bundle

@58414e9
by relatedbrelatedb/relate409 stars
35
Run it yourself

Relate is an embedded openCypher graph database with a single-binary CLI. Use this skill when the user wants to run Cypher queries, manage a local graph database, or embed it from Python; its focused references cover vector search, geospatial, operations, and agent safety.

  • 18 files
  • 116.4 KB
  • Updated 2 days ago
  • GitHub
Use this Skill: https://skilld.dev/gh/relatedb/relate/bundle

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

referencesclient-cli.md

≈1.2k tokens on demand. Your agent reads this file only when SKILL.md points to it.

CLI feedback reports

Install

Nothing extra: both feedback commands are built into the relate binary in every bundle. In agent mode agent-feedback appears in relate --help (feedback is the person's own command and is not listed there).

The agent flow: draft, ask, send

A report is the person's decision, never yours. On friction in a session — a command that failed, a wrong expectation, a missing capability — follow these steps in this order:

  1. Draft. Write a concise report: the observed problem, the failed attempts it caused, and the improvement you suggest.
  2. Ask. Show the draft to the person and ask for explicit approval.
  3. Send — only after that approval — with the full structural invocation:
relate agent-feedback \
  --category unclear_error \
  --expected "the info command to print the version" \
  --observed "it printed help instead" \
  --command info

An unapproved draft is dropped, never sent. Do not use the person's one-liner (relate feedback "…"): that command is the person's own GitHub issue form, exists only in their interactive run, and sends nothing.

--category defaults to friction; the other values are unclear_error, unexpected_behavior, and missing_capability. --expected and --observed are required. agent-feedback prints one JSON outcome line on stdout: {"outcome": ..., "reason": ..., "request_id": ..., ...}. Exit codes: 0 when the report was sent or skipped (telemetry opted out), 1 when it was locally rejected, the send failed, or no grant is recorded, 2 for a usage error. A locally rejected report records no payload in the local receipt.

The grant comes first

The send carries no permission of its own: agent-feedback requires the person's live standing grant, and without one it refuses with not_enabled (exit 1) naming relate feedback enable. The grant is the person's act at a terminal — ask them to run:

relate feedback enable   # authorize agent reports for 30 days

The CLI also asks the person directly, one line, default yes (an empty answer or y/yes grants; any other answer declines): once on their first interactive feedback command, and on skill install and relate upgrade runs — once per 30-day window, so a lapsed or declined grant is asked about again near its end. An opted-out telemetry ladder suppresses the ask entirely. Never try to grant it yourself: the agent-mode command does not exist, and a piped enable is refused.

Agent-mode error output also points here: an agent run's diagnostics carry a one-line hint to this flow, exactly once per run.

Gotchas

  • Ask before every send. The recorded grant is consent for the channel, not approval of this draft; an unapproved draft is never sent.
  • Free text is validated and redacted before any send. Emails, URLs, host names, IP addresses, file paths, data filenames, secrets, Cypher snippets and pasted output are rejected whole (never rewritten); a port-less token in a known procedure namespace (apoc.…, gds.…, db.ontology.…) is an identifier, not a host name — a token with an explicit port is always assessed as a host; the one exception is a documentation link to this project's own docs or repository, by path (relatedb.ai/<path>, github.com/relatedb/relate/<path>), or to the Neo4j Cypher manual (neo4j.com/docs/). A bare host URL with no path (relatedb.ai, github.com/relatedb/relate) is refused. --expected and --observed are at most 255 bytes.
  • Optional fields are dropped, not refused. --command survives only for a visible command path; --error-code only for a code name in the CLI's exit table (fatal_error, usage_error, conflict, validation_error, timeout); --model only for a lowercase provider-prefixed label (at most 40 chars). The outcome's dropped_fields names what was removed.
  • Nothing is queued. A report is one synchronous POST with a 2-second deadline and no retries; --request-id lets a retry keep the same id (and the $insert_id dedup key) after failed or unknown.
  • The recorded destination is the consent. RELATE_FEEDBACK_ENDPOINT is resolved when the person grants (relate feedback enable) and recorded; a later environment change cannot divert it, and a different destination needs a new grant.
  • Local receipts — feedback-receipts.jsonl in the config directory ($XDG_CONFIG_HOME/relate/ or ~/.config/relate/) records every rejection and send attempt.

Full contract: docs/api/cli/spec.md (flag tables, mode gating, exit codes, environment, state files).

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No third-party reports yet.

Signed by skilld at 58414e9. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated 2 days ago
version
{
  "{ VERSION }": null
}

README badge

README badge for relatedb/relate/bundle