Value Connectors¶
Value connectors turn external channels into reusable LoopX control-plane inputs. The first shipped path focuses on public GitHub metadata because it is useful immediately, does not require private data, and can be run by users after installing LoopX locally.
Quick Start¶
Install LoopX from the repository checkout:
When you are testing directly from an uninstalled checkout, replace loopx
below with ./scripts/loopx so the command uses the checkout code instead of an
older local release on PATH.
Check connector starter availability:
Give a newly connected agent the read-first connector source map:
Check the X/browser connector profile:
Probe a public GitHub issue or PR without network access:
loopx value-connectors github-public-probe \
--url https://github.com/owner/repo/issues/1 \
--format json
Probe body-free public metadata:
loopx value-connectors github-public-probe \
--url https://github.com/owner/repo/issues/1 \
--fetch-metadata \
--format json
Monitor whether a public maintainer replied after an approved LoopX comment:
loopx value-connectors github-reply-monitor \
--issue-url https://github.com/owner/repo/issues/1 \
--after-comment-url https://github.com/owner/repo/issues/1#issuecomment-123 \
--fetch-metadata \
--format json
The probe is intentionally metadata-only. It does not read issue bodies,
comment bodies, timelines, raw provider payloads, auth material, or local paths,
and it cannot post comments, send messages, create accounts, or publish.
The reply monitor follows the same boundary: it only captures comment author,
association, timestamp, and URL metadata, then emits either
prepare_public_triage_note or wait_no_bump.
Ownership And Compatibility¶
value-connectors is the compatibility facade for existing connector commands
and packet schemas. It owns shared install checks, source mapping, and gated
call planning, but it does not own new user outcomes. Each profile declares the
outcome capability that serves callers; implementations move there one proven
profile at a time while these CLI commands stay stable.
The first completed migrations are the public GitHub probe/reply monitor and
the social_browser_x profile. GitHub implementation and protocol ownership
live under issue-fix; the social source, install, and content-ops trial
contracts live under content-ops. Existing loopx value-connectors ...
commands delegate to those providers.
Connector Profiles¶
| Connector | Outcome capability | Binding | User can run now | External write behavior |
|---|---|---|---|---|
github_public_channel |
issue-fix |
migrated | yes | none |
github_public_reply_monitor |
issue-fix |
migrated | yes | none |
content_ops_public_handle |
content-ops |
native | public-handle observation | none |
social_browser_x |
content-ops |
migrated | install-check, public-handle packet, and gated plan | exact profile/post/reply gate required |
agent_reach_ops_source_map |
content-ops |
mapped | loopx value-connectors source-map --connector agent_reach_ops_source_map --format json; profile note |
publish/audit record required for every external write |
finance_market_snapshot |
none | migrated to standalone extension | migration packet only; no Finance execution | none |
botmail_identity |
content-ops |
mapped | install-check only | exact send gate required |
community_channel |
content-ops |
mapped | install-check and plan | exact account/message gate required |
migrated means the implementation module is owned by the outcome capability.
native means the command already lived there. mapped records the intended
owner without pretending that the implementation has moved.
finance_market_snapshot is retained only as an upgrade migration id. Its
source-map, install-check, and legacy plan --connector-id packets point
agents to the independently packaged loopx-finance-value-discovery extension
and never perform Finance work. See the
migration packet. The old id must not be
used for new integrations.
Why This Is Not Just A Plan¶
The plan command is the safety layer, but github-public-probe is a real
starter connector. It lets a user convert public channel URLs into compact
LoopX metadata and then decide whether to monitor, draft a reply, request
approval, or stop.
social_browser_x is intentionally one step more gated. It depends on
ego-browser for a logged-in browser session, media uploads, profile maintenance,
posting, and reply monitoring, but LoopX still owns the reusable control-plane
packet:
- observe public handles as metadata-only source items;
- plan account/profile work before touching the browser;
- require exact approval for every public post, reply, image, link, and mention;
- record a money, cost, demand, or capability metric plus a kill condition;
- monitor replies as compact signals instead of copying raw timelines.
Example X public-handle packet:
loopx content-ops observe-public-handle \
--url https://x.com/loopxops \
--source-item-id source_x_loopx_public_handle \
--no-fetch \
--format json
Example gated X publish plan:
loopx value-connectors plan \
--connector-id social_browser_x \
--connector-kind browser_social_channel \
--channel "X public post via ego-browser" \
--stage external_write_request \
--target-ref "one approved LoopX post" \
--target-url https://x.com/loopxops \
--external-write-requested \
--money-metric "qualified workflow owner asks for LoopX setup help" \
--success-metric "one audit, demo, or setup request" \
--kill-condition "spam hiding, account-health degradation, or no workflow owner signal" \
--format json
Future connectors should follow the same sequence:
install-check -> metadata probe -> value connector plan -> approval gate -> host connector execution
LoopX owns the compact control packet and value metric. Host products or user connectors own account login, private reads, external sends, and production actions.
Agent-Reach Ops Source Map¶
loopx value-connectors source-map --format json gives a newly connected agent
the current read-first connector catalog without requiring it to read internal
docs. It includes implemented or field-proven source profiles such as public
GitHub metadata probes, GitHub reply monitors, content-ops public handles,
browser-backed X research, and Agent-Reach source routing. It also names
action-gated profiles such as botmail and
community replies so agents do not treat "can send" as "can freely read/write".
agent_reach_ops_source_map is one profile in that packet. Agent-Reach is used
as a source router: first run agent-reach doctor --json, then collect
read-only signals from available routes such as GitHub, public web/RSS, V2EX,
or Bilibili. LoopX stores compact evidence cards, maturity scores, the ops
brief, draft packet, publish/audit record, and monitor state.
This profile is intentionally source-first and action-gated. Broad posting discretion does not remove the need to record exact body, channel/account, time, source refs, and stop conditions. See the Agent-Reach ops source-map profile.