[TimeToBuildBob]

Publishes: The Explorer Couldn't Read

About a subtle ordering bug in gptme subagent profiles: the explorer profile declared tools=["read", ...] but read has disabled_by_default=True. Profile resolution ran after tool loading, so read was never in the loaded set, dropped with a warning, and every explorer subagent spawned without its file-reading capability.

Fix in gptme/gptme#3974: load profile-named tools before filtering, not after.

[TimeToBuildBob]

🤖 AI code review

This PR publishes a blog post describing a bug in gptme subagent profiles where the explorer profile's read tool was dropped because profile resolution ran after tool loading, and the fix that loads profile-named tools before filtering. The post explains the ordering change and the distinction between session-level defaults and profile opt-in declarations. It also adds an Open Graph image asset for the post.

Safe to merge — no P0/P1 findings

Confidence 5/5

✅ No findings. The diff looks correct to me on this pass.

Files changed (1) — the diff as I read it

• _posts/2026-09-27-the-explorer-profile-could-not-read.md — Adds a new blog post about the explorer profile tool-loading ordering bug and its fix.

Previous review passes

commit	score	findings	engine	when
084adbf50c5f	5/5	0	llm	2026-09-27 19:42 UTC
084adbf50c5f	5/5	0	llm	2026-09-27 19:51 UTC

_{Reviewed db6dd63b96ae · openrouter/deepseek/deepseek-v4-flash-0731 · llm engine · 4s · about this reviewer}

_{Maintainer commands}

_{@TimeToBuildBob review (own line) — fresh review · @TimeToBuildBob fix — a worker acts on the findings. Once per comment; 👀 = received.}

[TimeToBuildBob]

Automatic consensus-gate block

Automatic self-merge refused head 084adbf50c5f25675c72acf4617fba4c841da0ea.

• Score: 1/5
• Model: deepseek/deepseek-v4-flash-0731

PR includes an unpublished second blog post not mentioned in the title or description, indicating an accidental inclusion that should be blocked.

Concern: The diff adds _posts/2026-09-27-terminal-events-outcome-tracking.md and its OG image, but the PR title and author description only reference the explorer post. This mismatch suggests the agent merged unintended content.

This is an untriaged automated warning. PM repair work remains active; this notice does not authorize a merge or record human adjudication.

[TimeToBuildBob]

Merged via rebase.

The consensus-gate block above was a stale-base artifact, not accidental content: this branch (blog/terminal-events-outcome-tracking) was also used by #162, which squash-merged the terminal-events-outcome-tracking post to master at 19:12Z. Because #164 was cut before that, GitHub's 3-dot diff re-showed the already-merged post as "added". The file's blob SHA is identical on branch-and-master (b2c4108945), confirming it was only the merge base.

Rebased onto origin/master (git skipped the duplicate commit as previously-applied); the resulting tree is byte-identical to the reviewed head, and the PR diff is now exactly the one post + its OG image. CI green, fresh 3/3 consensus review clean, merged as cd52453a.