Skip to content

CHAINS.md

Latest commit

 

History

History
44 lines (30 loc) · 1.51 KB

CHAINS.md

File metadata and controls

44 lines (30 loc) · 1.51 KB
Raw

CareerOps mode chains (human-gated)

Declared skill pipelines over a board pack. Each step writes an mt_reports row; human confirm is required between steps. Never auto-apply / never auto-send / never invent experience.

Builtin chains

Id Steps
prep-pipeline evaluate → rank → interview
evaluate-interview evaluate → interview

CLI

# list
npx @telivity/careerops run-chain --list

# start (writes chain run metadata into the pack)
npx @telivity/careerops run-chain prep-pipeline --pack CareerOps_board_pack.json --role <role_id>

# after the agent writes mode output, attach it and wait for confirm
npx @telivity/careerops run-chain prep-pipeline --pack CareerOps_board_pack.json \
  --run <run_id> --report-file step.json

# human gate — required to advance
npx @telivity/careerops run-chain prep-pipeline --pack CareerOps_board_pack.json \
  --run <run_id> --confirm

--confirm without a prior --report-file / writeStepReport for the current step fails closed.

Skill / agent surface

  1. Load pack (schema_version ≥ 4; optional extensions.chain_runs).
  2. startChain → follow brief.instructions for mode evaluate (etc.).
  3. writeStepReport with materials-only body.
  4. Stop and ask the human to confirm.
  5. confirmStep({ confirm: true }) → next brief or done.

Helpers: web/lib/mode-chains.mjs

Out of scope

Autonomous apply, unsupervised parallel agents inventing bullets, skipping confirm gates.