blog: dogfood field report on outreach agent governance#592
Merged
Conversation
A real internal workflow report on putting Cycles in front of an autonomous outreach runner: research → triage → Gmail drafts → Slack review → Attio sync → reply detection. Documents the budget vs. action authority split, the $0 toolset budget = DENY pattern, real receipts (RELEASED, COMMITTED, 409 BUDGET_EXCEEDED), what broke (microcent underfunding, budget-mutation 500s, Slack webhook misconfig, Gmail OAuth setup), and the systemd timer layout running the loop without a laptop. Companion piece to the scalerX integration post. Inline cross-links to runtime authority, action-control, audit-trail byproducts, real-time budget alerts, policy drift, reserve-commit protocol, and agents-cross-cutting.
…ction-kind Round 1 reviewer fixes: - Intro restructured: DENY payoff (`409 BUDGET_EXCEEDED` on external send, draft kept for review, side effect blocked) now appears in the lede instead of the receipts section. - Action-kind clarification: `send-email`/`send-email-approved` are toolset names; the canonical action kind in the v0.1.26 registry is `message.email.send`. Avoids spec confusion. - "What broke" #2 reframed from "budget mutation broken" to "admin-plane mutation hardening" — runtime path (reserve, commit, release, deny) worked; mutation paths are the hardening item. - "What happened in the first live runs" compressed: cut 18-artifact table and standalone triage table; queue counts moved to prose. - Keywords block tightened to reviewer's focused set. Skipped: title alts (current 51 chars vs 67-char alternative), date shift (explicit author choice), description rewrite (reviewer's version was over the 150-160 char SEO limit).
…rding - Add a clarifying sentence near the `$0.0000` allocation paragraph: the dollar amount is acting as an action allowance, not a provider cost. Removes the "wait, email costs $0?" confusion. - "VM with systemd" → "VM under systemd" (more precise relationship). - "while being denied" → "while still being denied" (smoother). - "That is the useful split." → "That split is the point." Skipped: optional audit-table-to-prose cut (reviewer marked optional; the table is concrete and reinforces the operator-contract claim).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
409 BUDGET_EXCEEDED), what broke during rollout, and the systemd timer layout running the loop without a laptop.Review process completed
runtime-authority,budgets,action-control,operations,case-study,agents)409 BUDGET_EXCEEDED, reserve→commit→release, $0-toolset=DENY, USD_MICROCENTS all confirmed against corpusTest plan
npm run dev) — H1, frontmatter, tables, inline links renderNote for author
One ledger-arithmetic question worth eyeballing before publish: line ~145 shows total allocated
\$61.6000and remaining\$50.7843, withresearch-liveremaining\$48.6845. The implied historical-toolset remaining is ~$2.10 but theresearch+research-rerunbreakdown only sums to ~$0.50 — possibly other older toolsets not shown, or different snapshots. Numbers are real-data so left untouched; flag if you want a one-line clarification.