McRitchie Studio

← All Tasks

Sizing

Narrate PR reviewer activities

Archived

narrate-pr-reviewer-activities

Created

Jul 06, 01:00

Started

Jul 06, 01:08

Completed

Jul 06, 02:07

DevOps handoff

Type

Bug

Shape

backend

Worktree Slug

narrate-pr-reviewer-activities

Repositories

mcritchie-studio

Release Slug

—

Branch

feat/narrate-pr-reviewer-activities

Pull Request

https://github.com/amcritchie/mcritchie-studio/pull/409

Local URL

—

QA URL

—

Production URL

—

observability agent-ops

Acceptance Criteria

Reviewer prompt tells each reviewer to narrate its activity
Each spawned reviewer becomes its own soul-attributed activity
Review verdict and scout-report recording remain unchanged

Expected Test Plan

[unit] reviewer_prompt includes agent-activity start/end lines
[integration] reviewer flow yields a narrated activity

Checks Run

[unit] bin/rails test test/commands/pr_review_command_test.rb -n /test_reviewer_prompt_instructs_each_reviewer_to_narrate_as_its_soul/
[integration] bin/rails test test/commands/pr_review_command_test.rb -n /test_run_mode_delivers_narration_instructions_to_each_spawned_reviewer/
[full-suite@1cd322bacf25a7461aed7469284667a5e1eb3ad3] bin/rails test green
[rubocop@1cd322bacf25a7461aed7469284667a5e1eb3ad3] bin/rubocop clean

Agent Context

Observability epic, Workstream B1 (foundation PR #407 merged 2026-07-06). Root cause: AgentAction capture is a per-session PostToolUse hook pinned by one .agents/sessions/<session_id>.open-activity marker. bin/pr-review launches PRIMARY+LIGHT reviewers as separate codex exec subprocesses (Process.spawn, bin/pr-review:330) inside one Bash call, and reviewer_prompt (bin/pr-review:247-292) only asks for bin/devops-cycle --record-scout-report — it never tells reviewers to run bin/agent-activity, so their work opens no activity and vanishes. Fix: add the bin/agent-activity start (--category Verify --agent <soul> --reason 'review: <scope>') and end (--outcome '<verdict>') that pr-review-sop.md:101-105 already prescribes to reviewer_prompt, so each reviewer becomes its own soul-attributed, gradeable activity. NOT trying to nest under the supervisor (harness-level limit per agent_activity.rb:22-26). Scope: bin/pr-review only; do not change verdict logic or scout-report recording.

Stage Timeline

Who handled each stage, the time it took (measured), and the model / tokens / cost reported (best-effort) — plus who's on it right now. — means the agent didn't report that metric.

Sizing Avi · PO SMALL Dev SMALL Actual XL ≠ forecast

Created → Designed

L
Lugia

Model

—

Duration

—

Tokens

—

Cost

—

Completed Jul 06, 01:00 · 2 days ago

api
Designed → Building

L
Lugia

Model

claude-fable-5

Duration

8 minutes

Tokens

1,002,478

Cost

~$7.87

Started Jul 06, 01:00

Completed Jul 06, 01:08 · 2 days ago

cli
Building → Submitted

L
Lugia

Model

claude-fable-5

Duration

11 minutes

Tokens

862,815

Cost

~$2.19

Started Jul 06, 01:08

Completed Jul 06, 01:19 · 2 days ago

cli
Submitted → Reviewed Review Activity

C
Carl primary

A
Alex light

Model

claude-opus-4-8

Duration

11 minutes

Tokens

1,553,909

Cost

~$1.93

Started Jul 06, 01:19

Completed Jul 06, 01:30 · 2 days ago

cli
Reviewed → Assembled

S
Steffon

Model

claude-opus-4-8

Duration

22 minutes

Tokens

13,138,667

Cost

~$9.58

Started Jul 06, 01:30

Completed Jul 06, 01:52 · 2 days ago
Assembled → Shipped

A
Avi

Model

claude-fable-5

Duration

15 minutes

Tokens

—

Cost

—

Started Jul 06, 01:52

Completed Jul 06, 02:07 · 2 days ago
Shipped → Archived

—

Model

—

Duration

about 14 hours

Tokens

—

Cost

—

Started Jul 06, 02:07

Completed Jul 06, 15:43 · 2 days ago