huly-api（huly API 搜索）

Huly API

Overview

Use this skill whenever swarm agents need durable communication in Huly:

• tracker tasks board (issues)
• chunter channels (chat messages)
• document teamspaces (mission docs)

Every autonomous mission should update all three surfaces. Default destinations for this workspace:

• Tracker issues board: DefaultProject (https://huly.proompteng.ai/workbench/proompteng/tracker/tracker%3Aproject%3ADefaultProject/issues)
• Documents teamspace: PROOMPTENG
• Chat channel URL: https://huly.proompteng.ai/workbench/proompteng/chunter/chunter%3Aspace%3AGeneral%7Cchunter%3Aclass%3AChannel?message

Required Environment

Load credentials from the huly-api Kubernetes secret via AgentRun spec.secrets.

• HULY_API_BASE_URL (should target transactor, for example http://transactor.huly.svc.cluster.local)
• HULY_BASE_URL (optional human-facing base URL, often http://front.huly.svc.cluster.local)
• HULY_API_TOKEN (bearer token)
• Optional:

- HULY_WORKSPACE (workspace slug for operator context) - HULY_WORKSPACE_ID (workspace UUID; helper can also infer from JWT) - HULY_PROJECT (task board project identifier/name/id) - HULY_TEAMSPACE (document teamspace name/id) - HULY_CHANNEL (chat channel name/id) - per-agent tokens: HULY_API_TOKEN_<SWARM_AGENT_IDENTITY> or HULY_API_TOKEN_<SWARM_AGENT_WORKER_ID>

Helper Script

Use scripts/huly-api.py.

When Jangar runs this helper from the bundled agent image, the script is copied under /root/.codex/skills/huly-api/scripts/huly-api.py. Local repo checkouts still use skills/huly-api/scripts/huly-api.py, and HULY_API_SCRIPT_PATH can override either layout for debugging. Jangar runner images also copy the helper under /app/skills/huly-api/scripts/huly-api.py because AgentRun auth mounts shadow /root/.codex at runtime.

Mandatory channel interaction loop

For every mission stage, run this in order:

1. Read recent channel history:

python3 skills/huly-api/scripts/huly-api.py \
  --operation list-channel-messages \
  --worker-id "${SWARM_AGENT_WORKER_ID}" \
  --worker-identity "${SWARM_AGENT_IDENTITY}" \
  --require-worker-token \
  --channel "${ACTIVE_HULY_CHANNEL}" \
  --limit 30

1. React to at least one relevant teammate message by posting a natural direct reply with decision impact in-thread:

python3 skills/huly-api/scripts/huly-api.py \
  --operation post-channel-message \
  --worker-id "${SWARM_AGENT_WORKER_ID}" \
  --worker-identity "${SWARM_AGENT_IDENTITY}" \
  --require-worker-token \
  --channel "${ACTIVE_HULY_CHANNEL}" \
  --reply-to-message-id "${RELEVANT_MESSAGE_ID}" \
  --message "Thanks for the context. I picked this up and I am implementing X next."

1. Post a worker-authored stage update after completing the mission action:

python3 skills/huly-api/scripts/huly-api.py \
  --operation post-channel-message \
  --worker-id "${SWARM_AGENT_WORKER_ID}" \
  --worker-identity "${SWARM_AGENT_IDENTITY}" \
  --require-worker-token \
  --channel "${ACTIVE_HULY_CHANNEL}" \
  --message "${OWNER_UPDATE_MESSAGE}"

Mission-level sync (recommended)

python3 skills/huly-api/scripts/huly-api.py \
  --operation upsert-mission \
  --worker-id "${SWARM_AGENT_WORKER_ID}" \
  --worker-identity "${SWARM_AGENT_IDENTITY}" \
  --mission-id jangar-discover-20260302 \
  --title "Jangar discover cycle" \
  --summary "Top platform risks and next actions" \
  --details "Includes evidence, risk deltas, and PR links" \
  --stage discover \
  --status running \
  --swarm-agent-worker-id "${SWARM_AGENT_WORKER_ID}" \
  --swarm-agent-identity "${SWARM_AGENT_IDENTITY}" \
  --message "${OWNER_UPDATE_MESSAGE}" \
  --project "DefaultProject" \
  --teamspace "PROOMPTENG" \
  --channel "general"

This creates/updates:

• one task-board issue
• one mission document
• one channel message

Single-surface operations

python3 skills/huly-api/scripts/huly-api.py --operation create-issue --title "..." --mission-id "..."
python3 skills/huly-api/scripts/huly-api.py --operation create-document --title "..." --mission-id "..."
python3 skills/huly-api/scripts/huly-api.py --operation repair-project-issues --project "DefaultProject" --limit 500
python3 skills/huly-api/scripts/huly-api.py --operation repair-project-issues --project "DefaultProject" --limit 500 --fill-empty-issue-descriptions
python3 skills/huly-api/scripts/huly-api.py --operation dedupe-project-mission-issues --project "DefaultProject" --limit 1000 --dry-run
python3 skills/huly-api/scripts/huly-api.py --operation repair-teamspace-documents --teamspace "PROOMPTENG" --limit 500
python3 skills/huly-api/scripts/huly-api.py --operation post-channel-message --message "..."
python3 skills/huly-api/scripts/huly-api.py --operation post-channel-message --channel "general" --reply-to-message-id "<parent-message-id>" --message "In-thread reply"
python3 skills/huly-api/scripts/huly-api.py --operation list-channel-messages --channel "general" --limit 30
python3 skills/huly-api/scripts/huly-api.py --operation account-info --worker-id "${SWARM_AGENT_WORKER_ID}" --require-worker-token
python3 skills/huly-api/scripts/huly-api.py --operation verify-chat-access --worker-id "${SWARM_AGENT_WORKER_ID}" --worker-identity "${SWARM_AGENT_IDENTITY}" --require-worker-token --channel "general" --message "<worker-authored access check based on the actual requirement>"

Raw HTTP mode (debug)

python3 skills/huly-api/scripts/huly-api.py --operation http --method GET --path /api/v1/account/<workspace-id>

Operating Rules

• Every mission stage must publish artifacts to all three Huly modules: tasks, channels, and docs.
• Each swarm agent should authenticate with its own Huly organization account (no shared actor credentials).
• Always target Tracker DefaultProject issues and Docs PROOMPTENG teamspace unless a run explicitly overrides them.
• Resolve ACTIVE_HULY_CHANNEL dynamically from runtime context (prefer swarmRequirementChannel, then HULY_CHANNEL, then default).
• Read channel history before taking action, then react to relevant teammate messages with explicit in-thread follow-up replies using --reply-to-message-id.
• Use --require-worker-token for dedicated-account checks so shared fallback tokens are rejected.
• For strict identity mapping, set per-worker HULY_EXPECTED_ACTOR_ID_<SWARM_AGENT_IDENTITY> and use --require-expected-actor-id.
• Run verify-chat-access before autonomous delivery stages with a worker-authored --message.
• Include swarmAgentWorkerId and swarmAgentIdentity in issue/document artifacts.
• Keep channel messages human-readable by default; avoid appending metadata blocks.
• Use actual run context in worker-authored updates; avoid canned acknowledgements or stock status templates.

- If you need metadata in chat for a specific audit/debug flow, use --append-channel-metadata. - upsert-mission automatically infers these fields from --worker-id/--worker-identity (or SWARM_AGENT_WORKER_ID/SWARM_AGENT_IDENTITY) if explicit --swarm-agent-worker-id/ --swarm-agent-identity flags are not provided.

• Keep channel updates concise, written by the worker, and linked to issue/doc/PR/deploy evidence.
• Issue descriptions must be long-form and knowledge-base quality. The helper now stores issue descriptions through collaborator refs (not raw inline markdown) so tracker descriptions reliably render and persist.
• upsert-mission now reuses existing mission issues by mission id prefix ([mission:<id>]) to avoid duplicate ticket creation.
• Use repair-project-issues to migrate legacy issues that still have inline markdown/plain text in Issue.description.
• Use repair-project-issues --fill-empty-issue-descriptions when historical issues have truly empty description fields.
• Use dedupe-project-mission-issues to move older duplicate mission issues to Canceled while keeping the most recent active issue.
• Document content must be long-form and knowledge-base quality. The helper now stores mission document bodies through collaborator content refs (not raw inline markdown) so Huly editor/collaborator consistently loads and renders content.
• Use repair-teamspace-documents to migrate legacy docs that still have inline markdown in Document.content.
• If Huly auth fails, stop and return a clear unblock request instead of silently continuing.