docs: example agent communications with raw API responses

[evansenter]

Summary

• Adds docs/EXAMPLE_COMMS.md with real conversations extracted from the live event bus
• Curated examples organized by communication pattern: coordination, debugging, cross-repo collaboration, knowledge sharing, infrastructure setup, social
• Raw JSON API responses section for agent researchers showing actual get_events, list_sessions, and list_channels formats
• Includes a complete raw multi-turn conversation thread with annotations on emergent conventions (asymmetric channel targeting, no threading primitive, Re: prefix conventions)

Test plan

• Verify markdown renders correctly on GitHub
• Confirm raw JSON examples are valid JSON
• Check all event IDs and timestamps match real data

🤖 Generated with Claude Code

[claude]

Prompt: evansenter/dotfiles/.../claude-review.md

Code Review

Summary

Adds docs/EXAMPLE_COMMS.md — a curated collection of real inter-session conversations extracted from the live event bus, organized by communication pattern (coordination, debugging, cross-repo collaboration, knowledge sharing, infrastructure, social), plus raw JSON API response examples for agent researchers.

Issues Found

Critical

None

Important

None

Suggestions

• docs/EXAMPLE_COMMS.md:31,53,244,330,358 - Several curated examples reference the pre-rename repo name claude-event-bus, and lines 251, 364, 752 reference claude-session-analytics. Since the curated sections are described as "lightly edited for readability" (line 857), consider adding a brief note (e.g., near the top or in the "How This Document Was Produced" section) that some events use pre-rename names (claude-event-bus → agent-event-bus, claude-session-analytics → agent-session-analytics) to avoid confusion for readers who only know the current names.
• docs/EXAMPLE_COMMS.md (multiple locations) - Real Tailscale hostnames (speck-vm.tailac7b3c.ts.net) and GCP-style machine identifiers (speck-vm.europe-west2-c.c.speck-ai.internal) appear throughout the document. While Tailscale URLs aren't publicly routable, these do expose internal infrastructure naming in a public repo. Worth a deliberate decision on whether to keep or redact these.
• docs/EXAMPLE_COMMS.md:829 - The "How This Document Was Produced" section uses Python-style function call syntax (get_events(cursor=0, limit=50, ...)) but get_events is an MCP tool, not a Python function. A brief note clarifying this is pseudocode showing MCP tool parameters (not a literal Python API) would help readers unfamiliar with the project.

Verdict

REQUEST_CHANGES - Three suggestions: add note about pre-rename repo names in historical data, consider redacting infrastructure hostnames, and clarify MCP tool pseudocode in the methodology section.

---

Automated review by Claude Code

[claude]

Prompt: evansenter/dotfiles/.../claude-review.md

Code Review

Summary

Adds docs/EXAMPLE_COMMS.md — a 989-line curated collection of real inter-session conversations extracted from the live event bus, organized by communication pattern (debugging, cross-repo negotiation, knowledge propagation, etc.), plus raw JSON API response examples and a methodology section.

Issues Found

Critical

None

Important

• docs/EXAMPLE_COMMS.md:937 vs docs/EXAMPLE_COMMS.md:848-871 - The "How This Document Was Produced" section (line 937) explicitly lists ci_completed, task_started, task_completed, wip_checkpoint, and wip_cleared as excluded event types ("routine, not communicative"). However, the "Status Broadcasts" section (lines 848-871) includes examples of exactly these event types (task_started, ci_completed, feedback_addressed, task_completed, wip_checkpoint, wip_cleared). This is a direct contradiction — the methodology claims these were filtered out, but the document showcases them. Either remove the Status Broadcasts / WIP checkpointing sections, or update the excluded types list to reflect what's actually included.

Suggestions

• docs/EXAMPLE_COMMS.md:348,543,648,672,740-741 - Several curated (non-raw-JSON) sections reference the pre-rename repo name claude-event-bus, and lines 172, 677, 746 reference claude-session-analytics. The raw JSON sections are clearly labeled as preserved verbatim, which is fine. But the curated sections are described as "lightly edited for readability" (line 3), which creates ambiguity — readers unfamiliar with the rename may not realize claude-event-bus and agent-event-bus are the same repo. Consider adding a brief note near the top (e.g., "Some historical events reference pre-rename names: claude-event-bus → agent-event-bus, claude-session-analytics → agent-session-analytics").
• docs/EXAMPLE_COMMS.md:52,96,541,547,807,814,822,824,836-837 - Real infrastructure identifiers appear throughout: GCP-style internal hostnames (speck-vm.europe-west2-c.c.speck-ai.internal), Tailscale URLs with network ID (speck-vm.tailac7b3c.ts.net), home directory paths (/home/evansenter/Documents/projects/...). While Tailscale URLs aren't publicly routable, these expose internal infrastructure naming conventions in a public repo. This should be a deliberate decision — if intentional, no action needed.
• docs/EXAMPLE_COMMS.md:788-838 - The "Infrastructure Setup" section substantially overlaps with content already presented in "Performance debugging across machines" (lines 228-236, same server setup conversation) and "Deploy coordination" (lines 521-535, same pull-and-restart pattern). The server setup on speck-vm appears in both places nearly verbatim. Consider either removing the duplicates from Infrastructure Setup (referencing earlier sections) or consolidating so each conversation appears once.
• docs/EXAMPLE_COMMS.md:943-961 - The methodology section uses Python function call syntax (result = get_events(cursor=0, limit=50, ...)) but get_events is an MCP tool, not a Python function. Since the document is aimed at people "building on the event bus or studying the data model" (line 9), using Python syntax for MCP tool invocations could mislead readers into thinking there's a Python SDK. A brief note (e.g., "pseudocode showing MCP tool parameters") or using the MCP tool call format would be clearer.

Verdict

REQUEST_CHANGES - One important issue (methodology contradicts document content re: excluded event types) and four suggestions (pre-rename name note, infrastructure identifiers decision, duplicate content, MCP pseudocode clarification).

---

Automated review by Claude Code

[claude]

Prompt: evansenter/dotfiles/.../claude-review.md

Code Review

Summary

Adds docs/EXAMPLE_COMMS.md — a 973-line curated collection of real inter-session conversations extracted from the live event bus, organized by communication pattern (debugging, cross-repo negotiation, planning/delegation, knowledge propagation, social), plus raw JSON API response examples and a methodology section. The third commit addresses all feedback from two prior review rounds.

Issues Found

Critical

None

Important

None

Suggestions

• docs/EXAMPLE_COMMS.md:922,926 - feedback_addressed appears in both the "communication event types" extraction list (line 922) and the supplementary Status Broadcasts note (line 926, "these were manually selected, not part of the bulk extraction above"). This dual classification is slightly contradictory — if feedback_addressed is a communication type extracted in bulk, it doesn't need to also be called out as manually selected supplementary content. Consider removing it from one list or the other to avoid ambiguity about how it was sourced.

Previously Addressed (Filtered)

• docs/EXAMPLE_COMMS.md:5 - Pre-rename repo name note (Implemented: added > **Note on names:** block at top of document)
• docs/EXAMPLE_COMMS.md (multiple locations) - Infrastructure identifier redaction (Implemented: hostnames replaced with server-vm.internal, server-vm.tailnet.ts.net, paths with /home/user/...)
• docs/EXAMPLE_COMMS.md:926 - Methodology contradiction re: excluded event types (Implemented: added clarifying note that Status Broadcasts examples are "supplementary reference — manually selected, not part of the bulk extraction")
• docs/EXAMPLE_COMMS.md:798 - Duplicate Infrastructure Setup content (Implemented: consolidated with cross-reference "See also the server setup portion of...")
• docs/EXAMPLE_COMMS.md:934-951 - Python pseudocode for MCP tools (Implemented: replaced with MCP tool parameter format using colons instead of =)

5 items from prior feedback rounds were not re-raised.

Verdict

REQUEST_CHANGES - One suggestion regarding feedback_addressed appearing in two contradictory classification lists.

---

Automated review by Claude Code

[evansenter]

Feedback Addressed

Implemented

• [Important] Methodology contradiction — Updated excluded types list to note Status Broadcasts are supplementary reference, not part of the bulk communication extraction
• [Suggestion] Pre-rename repo names — Added note at top: claude-event-bus → agent-event-bus, claude-session-analytics → agent-session-analytics
• [Suggestion] Infrastructure identifiers — Redacted all Tailscale URLs, GCP hostnames, and home directory paths to generic names (server-vm.internal, server-vm.tailnet.ts.net, ~/projects/)
• [Suggestion] Duplicate Infrastructure Setup — Removed duplicate server setup conversation, added cross-reference to Performance Debugging section
• [Suggestion] MCP pseudocode — Replaced Python function call syntax with actual MCP tool parameter format