Context Layer — origins and iterations
From Family Shapes project docs to reusable scaffolds, human review, dashboards and the Context Layer plugin, with examples across projects.
Imported markdown
Context Layer — origins and iterations
Context Layer grew out of Maggie Lerman’s project work. The recovered sequence begins with Family Shapes documentation, continues through a reusable project-docs scaffold and its workflow extensions, and reaches the named plugin in May 2026. The dates below distinguish a surviving implementation, a Git snapshot and a conversation reporting work. They do not assign a precise invention date where the record cannot support one.
The lineage
| Date | Iteration | What changed and why | Evidence |
|---|---|---|---|
| October 25, 2025 | Family Shapes project documentation | Numbered project plans; active, completed, backlog and archived folders; project IDs referenced in roadmap and changelog. Permissions, sharing and tree-builder work each had an owning record. | Family Shapes commit 76d00e48bb4675bdaebf99b995a32d15d55832c3, DOCS/PROJECTS/README.md. |
| By February 6, 2026 | docs-system-scaffold | Reusable DOCS structure, VitePress and manifest/link/frontmatter tooling. The skill explicitly says it is “modeled on the Family Shapes DOCS structure.” | Preserved session 019c3468-afb3-7d20-a17b-2bc68f3f28a7, records 11, 132 and 195. |
| February 16 | Docs-site automation default | Maggie asks to put automatic docs regeneration into the skill for future repositories. This is a skill-default request, not proof that every existing repository was retrofitted. | Session 019c6780-c504-7bb0-9d51-e4a2c64a734a, records 8 and 106. |
| March 10 | docs-project-lifecycle-statuses | Adds in-review for implemented work awaiting Maggie’s walkthrough and blocked for work needing her input. Applied to Shopify Headless, then folded into the standard scaffold; the standalone extension becomes a retrofit. | Session 019cd907-2b3d-7023-8b9d-26d5e900506a, records 7, 406 and 599. |
| March 21 | First reviewed Git-library snapshot | The skills library records the reusable scaffold and lifecycle companion. This snapshot postdates the February usage; it is not the skill’s creation date. | codex-skills commit 3f430f2. |
| April 5–8 | Generated HTML project dashboard | Maggie asks for a visible project/status view, checks that Git remains authoritative, and chooses HTML. The dashboard becomes an optional scaffold module; docs-project-dashboard-legacy remains for retrofit/backfill. | Session 019d5d58-eed4-7651-a3dd-675304c320f8, records 202, 231 and 385. |
| April 12 | MAGGIE TODO and a separate v2 experiment | Manual tests, approvals and missing input gain literal callouts. A separate repo-pm-orchestrator plugin / repo-pm-scaffold-v2 explores programs, dotted numbering and a multi-page dashboard. Maggie explicitly asks to protect the existing critical scaffold. | Session 019d81c9-a99d-73d1-95dc-11483e426586, records 6, 160, 243, 287 and 497. Implementation is reported in the conversation; the prototype’s original directory was not recovered. Its direct incorporation into Context Layer is unproven. |
| By May 12 | ML-product-operating-system-scaffold | The expanded skill coordinates intent, project state, agent instructions, evidence, checkpoints, review and launch memory; dashboard, docs site and product docs are still optional. | codex-skills commit 1893aa9, skills/ML-product-operating-system-scaffold/SKILL.md. The naming change may predate this snapshot. |
| May 13 | Context Layer named and packaged | Focused trigger skills solve recurring workflow discovery. Setup detail moves into references; DOCS and dashboard become standard; the template moves to the projects root; parent–child rules are codified. The existing scaffold is preserved as a rollback copy. | Evaluation session 019e2162-f99e-7270-afd1-6124d223b55a, records 181, 191, 249, 264, 310, 588, 675 and 1256. |
| May 15 | Versioned package snapshot | The skills library contains the Context Layer package and ten skills; the older Product Operating System skill survives as compatibility/history. ML- prefix removal is catalog naming history, not a separate new architecture. | codex-skills commit 46774fd. |
| June 11 | Dashboard backfill | A focused skill helps refresh existing installations instead of making each repository rediscover the dashboard migration. The reviewed package now contains eleven skills. | codex-skills commit ea5bc5b. |
The reviewed plugin manifest remains 0.1.0. The table describes recovered iterations and experiments, not a fabricated series of semantic version releases. The surviving docs-system-scaffold compatibility alias points to Product Operating System Scaffold; the historical lifecycle and dashboard aliases explain how older repositories could receive selected additions without repeating the entire setup.
The decisions in Maggie’s own words
These are selected excerpts from visible user messages, with original wording retained. They are not full public transcripts.
March 10 — implementation versus acceptance: “do not set hte project to completed when you work this way, though. you need to set it as in-review” (lifecycle session, record 7).
April 5 — one source of truth: “great, and what is the source of truth for the dashabord? its git based right?” (dashboard session, record 202).
May 13 — protect the working system: “i want to ensure that nothing gets lost so reliability matters a great deal” (evaluation session, record 181).
May 13 — why focused triggers: “sometimes the agent forgets when i say something like create a review board” (evaluation session, record 191).
June 24 — portability: “the foundation of my process is context-layer” (Cursor architecture session 019ef95e-8637-7641-9e17-3a4160b60748, record 234).
Examples beyond the portfolio
Family Shapes: a home for each workstream
The October 2025 structure gives permissions, privacy, tree-builder UX and union-node planning stable project identities. The February skill then names this structure as its model. That is direct lineage from a project’s internal practice to reusable tooling. It does not mean the October repository already had the eventual eleven-skill plugin.
Shopify Headless: autonomous work with a human review boundary
The March request distinguishes “the agent has implemented it” from “Maggie has reviewed it.” The resulting skill creates the two missing lanes, updates repository instructions, project templates and indexes, and passes link/manifest validation. The conversation subsequently records integrating those defaults into the reusable scaffold. The demonstrated result is a durable acceptance boundary and a reusable workflow; no throughput increase was measured.
Notion dashboard automation: the plan survives the conversation
The February 6 source records a real scaffold application, then Maggie asks to preserve an entire Notion/Observable dashboard build plan. The assistant stores it as Project 1001 in DOCS/PROJECTS/active. The useful result is modest and concrete: the next session has an owning plan to read. This is an early use of the scaffold, not evidence that the proposed Notion/Observable product was fully delivered.
Maggie Todo: a smaller system for human follow-ups
On May 23, Maggie asks for a running personal and cross-project todo repository without another hosted website. The recorded implementation provides Markdown todo sources, a generated local dashboard, a todo index and MAGGIE TODO support. An initial batch links a conversation, the Shopify project intake, the implementation of image evidence and a screenshot intake that depends on that implementation. The subsequent source records a pushed private repository. This example concerns the mechanism and dependency record; personal todo contents are not published here.
Frill: a handoff with state to reconcile
The September 24 audit reconciles 16 existing numbered project documents with requests, UX records, Git and retained release/provider evidence. It identifies resolved launch blockers that old checklists still describe as open, differentiates released work from earlier local-only notes, and carries forward Maggie’s explicit builder pause. Its outputs are a request ledger, action register, project/source inventories and a fresh-agent brief.
The audit also finds drift inside the system: duplicate project ID 1002, mixed lifecycle labels and historical plans that should no longer drive implementation. A subsequent lifecycle report resolves several of those findings while retaining the older evidence. That is a useful limit to keep visible: the system enables reconciliation, but still needs maintenance. The recovered report includes later addenda; its changing outstanding-action count is not used as a frozen September 24 metric.
Source: frill-hydrogen, DOCS/evidence/active/context-layer-audits/20260924-thread-handoff/report.md and fresh-agent-brief.md, inspected October 11. This page publishes the workflow findings rather than the full operational/customer record. These are historical examples, not claims about Frill’s current release or pause state.
Hyphenomenon: linking evidence to publication
Hyphenomenon’s project records connect intake artifacts, validation, preview, production application and live readback. The first standalone Context Layer page is tracked in Project 1164, with the prior portfolio chat, plugin manifest and skills library preserved as separate linked records. This expansion adds the source lineage to that same record. It demonstrates the practice on the site presenting the work.
A qualification to the portfolio story
The October 11 diagnosis concerns the active redesign checkout and the absence of the maintained standard layer there. A recovered May 13 conversation (019e22e9-8f67-7911-81bb-421c8289a44a, completion record 239) reports a lightweight scaffold in a different portfolio worktree. It included lowercase docs, lifecycle records and tooling, with dashboard, docs site, product pack and review board switched off. This evidence does not establish that the setup was merged into or maintained in the later redesign checkout. The accurate story is about continuity and adoption across the working repository, not an absolute claim that no portfolio worktree ever contained scaffold files.
Portfolio project · October morning conversation.
Native context and model adherence
In the May 31 source conversation (019e7e3a-5aed-7771-a15f-c33401ba0a2a), Maggie asks whether native Codex features are catching up with her custom context infrastructure. In June she calls Context Layer foundational while evaluating Cursor. In July, the compliance-playground investigation examines a model’s failure to follow installed governance. The existence of instructions and evidence is not proof an agent will use them correctly.
The October portfolio account supplies Maggie’s current assessment: even with richer native context support and gpt-6-astra high, repository state still needs explicit ownership and continuity. These records support the history, the mechanisms and specific uses. They do not provide a controlled comparison, a percentage productivity gain or proof that every failure came from one missing tool.
Source and publication boundaries
The source ledger retains original conversation IDs, selected record numbers and SHA-256 hashes, plus repository commit/path evidence. Full local histories, tool output, machine-specific paths, private todos and Frill operational details remain outside this public note. The October morning conversation is separately public under Maggie’s explicit permission; this research does not automatically publish every earlier conversation.
Context Layer project · Existing plugin manifest · Codex Skills Library.