Session Report — AI Dev Brain: public release & campaign launch

2026-07-15 → 2026-07-16 · repo valter-silva-au/ai-dev-brain (now public) · branch main

Summary

Goal: decide whether AI Dev Brain (adb) was safe to open-source, and if so, ship it with a social campaign built on brain.mp4. Outcome: a pre-publication audit found the code secret-clean but flagged real confidentiality problems — a full OpenClaw/Propella business plan reachable from published tags v1.7.0v1.9.0, pervasive Amazon fingerprints, and private-fleet over-share. On your call ("scrub every trace"), the tree was scrubbed (commit 8850c13, 31 files, +210 / −2362), verified green (build, vet, Go tests, extension 293/293), and republished as a fresh-history public repo (github.com/valter-silva-au/ai-dev-brain, single commit 4e74d41, no tags). The old repo was preserved private as ai-dev-brain-private. The campaign then went live: the brain.mp4 Gource render is posted on LinkedIn and this report captures its first-hour analytics plus a concrete next-steps plan.

Recap reflects available context + git facts; a long/compacted session may omit earlier steps. File lists are derived from git, not memory.

Timeline

  1. Audit — three parallel auditors swept the repo (secrets · Amazon/employer · personal/confidential), tree and git history.
  2. Verdict — secrets clean; but confidential OpenClaw/Propella doc lived in remote tags v1.7–v1.9, plus Amazon paths/Bedrock/internal codenames and private-fleet detail.
  3. Decisions (you) — scrub every Amazon trace; publish via fresh-history republish.
  4. Scrub — commit 8850c13: deleted stale/confidential docs + dead code + hive-era images, trimmed STORY.md, genericized employer/identity references across code, templates, docs, tests.
  5. Verifygo build/vet clean; affected Go tests pass; VS Code extension 293/293; full tracked-file sweep clean.
  6. Publish — renamed old private repo → ai-dev-brain-private (backup w/ full history+tags), created new public ai-dev-brain, pushed pristine single commit 4e74d41 over SSH.
  7. Campaign live (2026-07-16)brain.mp4 "This looks like generative art… how I now run a business" post published on LinkedIn; analytics pulled (screenshots + omnictx).

Campaign analytics — LinkedIn post (≈57 min in)

Source: your two analytics screenshots + omnictx capture 260716-012748 ("Post analytics | LinkedIn"). Post: urn:li:activity:7483316990091210752.

19
Impressions
13
Members reached
7
Video views
10s
Avg watch (1m10s total)
0
Followers from post
0
Profile views from post

In-network 53% · out-of-network 47%. Profile sidebar (7-day aggregate, not this post): 130 profile viewers, 1,040 post impressions. Engagement counts (reactions/comments/reposts) were below the screenshot fold and omnictx's CLI has no content-read (MCP-only), so they're not captured here.

Honest read (too early to judge — give it 24–48h): a <1h post at 19 impressions / 13 reached with 0 visible engagement means LinkedIn hasn't decided to push it — the first-hour engagement signal is the gate, and it wasn't pulled. The 10s average watch on the clip says the "generative art" hook works (people start it) but they don't finish. 0 followers / 0 profile-views-from-post = views aren't converting to relationship yet.

Recommendations — what to do next (content · strategy)

0. Positioning risk to decide consciously (not a code issue)

The post runs under your "Cloud Architect @ Amazon Web Services (AWS)" badge, framed as "how I now run a business" with #entrepreneurship / #buildinpublic — while you're also job-hunting (AWS ProServe / Anthropic / OpenAI, per the fleet). That's the same employer/IP/moonlighting tension flagged during the repo scrub, now public under your AWS identity. Your call, but decide it on purpose: (a) keep it (you've judged you're clear), (b) soften "run a business" → "how I work" to cut moonlighting optics, or (c) confirm with AWS if unsure. Nothing in the code forces this — it's a positioning choice.

1. First-hour engagement is the single biggest lever (do it in the golden window)

2. Put the repo link in a pinned comment, not the post body — this answers your "update the link" question

3. Repurpose the clip everywhere (the asset is inherently shareable)

4. Make it a series with a takeaway + CTA

5. Ladder the campaign toward the product

6. Land the click-throughs (quick win before more traffic)

Reality check on measurement

There is no LinkedIn analytics API — screenshot → omnictx (what you did) is the working loop for LinkedIn. X analytics need xurl + a funded API plan. The clip campaign's real dashboard is the content skill (tools/clip/content.jsonreports/content.html). This is TASK-00053 (social-perf tracking, in progress).

Artifacts produced

ArtifactWhat it is
github.com/valter-silva-au/ai-dev-brainThe shipped public open-source repo — fresh single-commit history (4e74d41), no tags, anonymously reachable (HTTP 200).
C:\Users\valte\adb-public-stagingPristine fresh-history snapshot that was pushed to the public repo (tracked files only, no old history/tags). Disposable now.
memory/ai-dev-brain-public-release-prep.mdDurable release record — the audit verdict, the tag-leak warning ("never make ai-dev-brain-private public"), and what shipped.
reports/session-report-2026-07-16-0930.htmlThis report.
LinkedIn post (activity 7483316990091210752)The live brain.mp4 campaign post ("how I now run a business").

Files edited

The scrub changeset — commit 8850c13 in the private clone (C:\Users\valte\Code\valter\repos\github.com\valter-silva-au\ai-dev-brain): 31 files, +210 / −2362. Notable:

FileChange
docs/architecture/hive-mind-design.mddeleted −1422 — private 8-bot fleet design, /home/valter paths, OpenClaw
docs/adb-llm-wiki-monorepo-assessment.mddeleted −311 — stale internal critique, leaked local path
prd.jsondeleted −164 — spec for the removed hive-mind (OpenClaw/Bedrock)
pkg/models/hive.godeleted −137 — dead code (zero importers)
docs/images/08–12*.pngdeleted — 5 hive-era campaign images
STORY.mdtrimmed −250 — removed Ch 9–10 (Hive Mind / migration); fixed image paths
templates/claude/rules/standing-dos-donts.md · task-context.mdgenericized — "Valter"→"project owner", ~/Code/amazon$ADB_HOME, Bedrock note
internal/observability/chat.goBedrock model default → sonnet; identity string genericized
issuesync select.go/sync_issues.go + tests, eventlog.go, launch.go/.ts, bootstrap*.go, hookengine_test.go, package.jsongenericized code.amazon.com/code.aws.dev/CRUX/Taskei/proserve and local usernames
.pre-commit-config.yamlremoved fleet-specific comments (roots-co / Hermes / .env.aws)

Key decisions

Tasks (adb — available)

Relevant open tickets this session's work touches (of 14 in-progress / 28 backlog):

TaskStatusRelevance
TASK-00053 — track personal-brand social performance (LI/X/IG)in progressThis report + the screenshot→omnictx loop is that tracking. No LinkedIn API exists.
TASK-00056 — 2-week daily short-form cadence (TikTok-first, Nina-drafted)backlogRepurpose the brain clip across channels; this is the vehicle.
TASK-00031 / 00033 — omnictx micro-SaaS ("private AI Dev Brain from a company link")backlogThe public adb + campaign is top-of-funnel for this product.
TASK-00029 — Content Agency v1 (productized recurring AI-content offer)backlogThe distribution muscle this campaign is building.

No adb task was created/closed this session; consider logging the release + campaign launch against a ticket (e.g. TASK-00053) so the knowledge is durable.

Sources & caveats

Fact sources used this run: git (commit 8850c13) adb tasks LinkedIn analytics (your screenshots) omnictx (corroboration) gstack (thin — 1 entry) transcript (wrong session).

Recap reflects available context + git facts; a long/compacted session may omit earlier steps. File lists are derived from git, not memory.