Skip to content

120. Record absorbed steers as agent-run receipts ​

Date: 2026-11-02

Status ​

Accepted

Companion decisions:

  • ADR 0119 (proposed): what a steer is. Steering is correct without this ADR.
  • ADR 0121 (proposed): the default and the opt-outs.

Context ​

ADR 0119 (proposed) lets a running cycle absorb entity updates, but the notification that prompted a steer still schedules a follow-up cycle. Unless something tells that cycle what was absorbed, it redoes the work.

ADR 0144 adds no Fullsend handled-state gate: handled state belongs in the harness's CEL or pre-script. ADR 0098 allows "an explicit per-harness receipt" for that purpose. A receipt that someone else can forge would make a harness skip work it still owes, so authorship is the core of the decision.

This ADR aims to let the next cycle's harness recognize absorbed updates without adding a runner skip and without trusting forgeable content.

Options ​

OptionWhy not chosen
Runner-level skip of absorbed recordsADR 0144 rules out a second handled-state gate
Post the receipt under the Actions job tokenEvery workflow that comments with GITHUB_TOKEN posts as the same github-actions[bot] account, so any of them can forge it
Post under the harness's Fullsend App identity; readers admit only registered bot_role authors (chosen)—

Decision ​

We adopt an agent-run receipt, written under the harness's Fullsend App identity, to let the next cycle's harness CEL or pre-script recognize the updates a steered cycle absorbed.

Fullsend adds no runner skip.

What is written ​

  • After a cycle with at least one acknowledged steer publishes validated output, the host writes one agent-run record for the cycle, in the ADR 0141 agent-run shape.
  • It names the absorbed record keys, the steer count, and the heads the steers offered. The heads are for audit only (see below).
  • The closed entity-context-snapshot v1agent-runs schema has no field for absorbed keys, so receipts need entity-context-snapshot v2.

Who may write it, and how a reader trusts it ​

  • The host posts the receipt as the harness's Fullsend App, the identity ADR 0107 recognizes with a bot_role. A workflow that posts with GITHUB_TOKEN posts as a different account and cannot produce a receipt a reader accepts.
  • A reader admits a receipt only when its author resolves to a registered Fullsend bot_role and the body has not been edited. Any other receipt is ignored, and the cycle runs.
  • The agent's own output is also posted under the App. Fullsend neutralizes its markers in every body it posts on the agent's behalf, so an agent copy of the receipt marker is inert.
  • No receipt is written for a cycle in which a credential that can post as that App was reachable from the sandbox, for example through a provider definition like the one in ADR 0114.

What a receipt does not prove ​

A steer never changes the cycle's platform-verified result revision (ADR 0119). An acknowledgement proves the message arrived, not that the agent acted on it. Consumers must not treat an acknowledged steer, or a head named in a receipt, as proof the agent worked on that newer head.

Reading it ​

The harness pre-script or CEL reads the receipt through the snapshot's agent_runs source, as ADR 0098 allows. Every doubt fails open: a missing, unrecognized, or unparseable receipt means the next cycle does the work.

Forge scope ​

  • GitHub: supported, posted as the harness's Fullsend App.
  • GitLab: not supported; ADR 0119 does not steer there.
  • Forgejo: not supported, for the same reason.
  • Jira: not applicable; Jira entities do not steer.

Consequences ​

  • A harness that reads receipts skips redundant follow-up cycles; one that does not still runs them, which stays correct.
  • Receipt trust rests on ADR 0107 bot identity and marker neutralization, not on a separate credential.
  • Receipts wait on entity-context-snapshot v2; until then every steered cycle is followed by a full cycle.
  • Auto-merge and approval gates keep comparing the verified result revision with live state; a receipt gives them no newer revision.

Open questions ​

  • Which shipped harnesses should read receipts first, and does a shared CEL helper belong in fullsend-ai/agents?