This page runs our 60 minutes today. It states where ShurIQ development actually stands as of this morning, what has to be decided in the call to unblock it, what Nicholas takes on, and how the context pack we built for the Manus agent last weekend becomes the way every one of us gets a worker agent that knows the real system.
2026-08-04, 13:00 to 14:00 EDT · Written for Nicholas Crabill · Google developer documentation style
Six blocks. The first half puts the sprint back on rails; the second half covers the agent-sync work and what it means for how you, Alex, and eventually the rest of the team work.
| Time | Block | Outcome we leave with |
|---|---|---|
13:00 | Where the plan actually stands | Shared, unromantic picture. No new work opened. |
13:07 | Unblock it: decisions made live | Repository name chosen. First three pull requests named. |
13:20 | The delegation process | Agreement on what a ready ticket is and how work reaches you. |
13:30 | What Nicholas takes | Three to five named items with acceptance checks. |
13:40 | The Manus correction, and why it matters | You have seen the pack and the failure it fixed. |
13:50 | Worker agents in sync with ShurIQ | Your pack scoped. Alex's named. The cadence set. |
13:58 | Close | Who does what before the next sync. |
Read the numbers in the next section together. The whole point of this block is that neither of us argues from memory. Every figure on this page was checked this morning against the GitHub API or against the files themselves.
The honest summary: sprint 1 is in specification, not execution, and it has been stalled since 2026-08-01 because the Black Flag application took the weekend. The work done before that is real and it holds. Nothing needs redoing.
One decision blocks nine others, and it is mine alone. We make it on the call rather than after it. See Decisions today for the queue, in blocking order.
What I want from this block is your pushback on the process itself before we run any work through it. The shape on paper: a story is not ready until it carries four fields, and you never start on one that does not.
Plus any open question addressed to a named person. A story missing a field is not ready, and the right answer to receiving one is to send it back.
Two habits that go with it. Answers land on the ticket, never in Slack. A blocked item carries a link to the thing blocking it, so "blocked" always has a target instead of a memory.
Full detail in the feature development process guide, sections on the seven stages and the definition of ready.
Working from the sprint-1 cut and the order of work. Proposals in Delegation below; we cut them to three to five items and put an acceptance check on each one before we leave the block.
An agent produced 23 video tutorials about our system. Almost none of them could be followed, because the agent had built its picture from starred and forked repositories on my personal GitHub account. We fixed it with a context pack rather than with better prompting. Walk through the pack in The agent-sync primitive.
The pack is the reusable part. The same anatomy, scoped to a person and a job, is how you get a worker agent that argues from what is true here, and how the DKR agency runs research overnight without producing work that has to be thrown away in the morning. Scope yours in this block.
Checked 2026-08-04 against the live GitHub API and against the files on disk. Anything not verified this morning is marked.
shuriq-lab, created 2026-07-28priority: nullThe blocking fact. The organization contains zero repositories and no repository name has been chosen. The only candidate on record is platform, recorded as a recommendation on 2026-07-28. That decision is mine alone, and every other item in the order of work sits behind it: the scaffold, the first pull request, continuous integration, the gates entering continuous integration, the registries becoming files under review, the route declarations, and the Jira project that would render issues out of the repository.
| Thing | State | Where |
|---|---|---|
| GitHub organization | Exists Team plan, 5 seats with 3 filled, two-factor required, owners-only repository creation | github.com/shuriq-lab |
| Monorepo seed folder | On disk README, one gate script, three registry drafts. Copies into the repository root as-is | Seed README |
route-gate.py | Written 451 lines, deterministic, no model calls, selftest 13 of 13. Proposed as the repository's first pull request so the check exists before the first route does | Gate section |
grammar-gate.py | Runs by hand Globs *.html only, so markdown is ungated by construction | Sprint plan, section 4 |
| Feature register | 120 rows 37 work today, 22 work with a person doing a step by hand, 61 are specified and unbuilt | Registry draft |
| Report Studio | Production The one conventional web application, outside the vault | shuriq-report-studio.pages.dev |
The full list runs to 23 items in the development setup state document. The ones that matter for today:
priority: null. Nothing has been ranked.One list. Steps 2 through 10 are all blocked on step 1.
| # | Step | Blocked on |
|---|---|---|
| 1 | Repository slug chosen | Jonny, today |
| 2 | Repository created, layout seeded from monorepo-seed/, dotfiles get a git home | Step 1 |
| 3 | route-gate.py merges as the first pull request; channels.yaml goes live through S-20 | Step 2 |
| 4 | Kickoff session runs on the walked client chain; five route declarations produced | Step 3 |
| 5 | The sprint-1 cut builds, run health first | Step 4 |
| 6 | Jira project mounts, once epics and stories exist to render | Step 5 |
| 7 | Registry adjudication completes the forms registry | Step 3 |
| 8 | Decision P-2 signs and the pathway modules become buildable | Jonny |
| 9 | Style coverage extends to markdown in continuous integration | Step 3 |
| 10 | The vertical intelligence block enters | Two questions, both Jonny's |
In blocking order. Every one of these is mine to make; the value of making them with you on the call is that you can tell me which answer costs you the least to build against.
The repository name. platform is the standing recommendation. Alternatives worth 60 seconds: shuriq, monorepo, stack. The constraint already decided: no hardcoded organization URLs anywhere, so a rename later costs nothing.
What lands in what order. Proposed: (1) the seed layout, (2) route-gate.py plus its selftest in Actions, (3) channels.yaml going live. Nicholas: does that give you enough to open a branch against, or is something missing from the scaffold?
Where work items live. TaskNotes was retired on 2026-07-25 and reinstated on 2026-07-28 as a provisional placeholder. Two stories are blocked on this entirely, and it decides where the review-and-assign action dispatches. Jira renders views once it exists, but the store question is separate.
Whether the tracker mounts before or after the first stories exist. The plan says after, on the argument that a tracker with nothing real in it teaches us nothing. Worth re-testing with you, because you are the one who has to work inside it.
Who ranks the 120 register rows, and against what. All 120 are priority: null today. This is KG's call on sequence, mine on scope. What we need from this call is the date it happens.
Whether a quality assurance function exists at all. Recorded as an open question. Right now I review and merge everything, which is the bottleneck we said we were removing.
Proposals, not assignments. We cut this to three to five items in the call and put an acceptance check on each. Your role on the record is front-end build and integration, building in the monorepo once it exists.
| Item | What it is | Branch | Acceptance |
|---|---|---|---|
| S-20 Run health readout | Ships first in the cut. Nothing blocks it. Wires channels.yaml live | feature/s20-run-health | KG accepts on the running product |
| S-02 Evidence drawer | Opens the evidence behind a figure. Blocked on nothing | feature/s02-evidence-drawer | KG accepts |
| S-06 Stack rank | The pipeline behind it already ships weekly | feature/s06-stack-rank | KG accepts |
Pull request review and merge stay with me for sprint 1, and so does every decision that enters the system. That is deliberate and it is also the thing we are trying to fix: the reason for the process above is that ready tickets and real gates are what make it safe to hand the merge button to somebody else. Say so on the call if you think a step of it is waste.
What we built for the Manus agent between 2026-08-01 and 2026-08-02, why it worked, and why it is the pattern for everyone's agents rather than a one-off repair.
An agent produced 23 video tutorials describing how our system works. Almost none of the instructions could be followed. The agent had read the starred and forked repositories on my personal GitHub account and treated them as the working stack.
| Repository named in the tutorials | Times | Actual role here |
|---|---|---|
graphify | 97 | None |
knowledge-base-server-demo | 78 | None |
obsidian-mcp | 61 | None |
SimpleMem | 46 | None |
mempalace | 42 | None |
NexAU | 24 | None |
Every instruction built on those names a tool that is not installed and not called here. The agent was capable, the prompting was fine, and the output was unusable, because the picture underneath it was wrong. Prompting harder would not have touched it.
We shipped the agent a context pack: a folder it reads before it writes anything. Six parts, and the shape is the reusable thing.
| Part | What it does | Why it is there |
|---|---|---|
| Ground truth | The verified condition of the system on a stated date: what runs, what is prototype, what is specified and unbuilt | One document wins every disagreement. Without a named authority, a conflict gets resolved by whichever document the agent read last |
| Technical brief | What each component does and how the three layers compose | Mechanism, so the agent can reason instead of pattern-match |
| Development setup state | GitHub, Jira, the monorepo, the gates, the roles, the order of work, with planned things labelled planned | The tense rule. This is where agents fabricate most: describing a plan as a running system |
| Concept space reader and 15 posts | How we argue, at what level of specificity, and the writing rules a program checks | Voice by demonstration. Rules alone do not carry it |
| The OKF bundle | The same knowledge as a portable container of markdown concepts, one profile, machine-readable | What the agent consumes and emits, so its output comes back in a shape we can ingest |
| Prompt 1 and Prompt 2 | Ingest and report back, then do the work | The pack is inert until something makes the agent prove it read it |
The part worth generalizing. Nothing about this is specific to Manus, to video tutorials, or to that agent's mistake. It is a way of shipping a person's or an agent's working picture of ShurIQ as a versioned artifact, with an authority, an export policy, and a test that it landed. That is a primitive for the DKR agency: every worker agent gets one, they are rebuilt on a cadence, and drift shows up as a failed ingestion report instead of as unusable work three weeks later.
The pack is a Layer 2 export by decision, not by convenience. Withheld: the scoring weights, formulas, and rubric internals; typed edge weights and confidences from the store; any client's raw material, scores, or name-linked findings; anything under the protected frontmatter namespaces; internal self-audit and methodology. Two tutorials and two sections of the format profile were held back, and each place says so where the removal happened.
This is what makes the pattern safe to hand to a third-party agent, and it is the same reason a teammate's pack can be generous where an external agent's cannot. The layer decision comes before the packaging, every time.
| Pack | For | Scoped to | Export layer |
|---|---|---|---|
| Nicholas's build agent | Front-end build and integration | The monorepo layout, the gates, the registries, the delivery grammar, the definition of ready, the running deploys | Internal, generous. No client raw material |
| Alex's design agent | Form rendering and design language | The form registry, the rendered report pages, the visual grammar, the design rubric | Internal. No scoring internals |
| Research agents, overnight | DKR agency operations | The operating loop, gap-finding, what is already covered, the writing rules | Whatever the destination allows |
| Third-party agents | Anything outside the team | Whatever the job needs and nothing else | Layer 2 only, on the Manus template exactly |
The open questions I want your answer on: what cadence rebuilds a pack, what triggers a rebuild between cadences, and whether the ingestion report gets reviewed by a person or checked by a program.
Sixteen documents, copied from the vault this morning and rendered here so nothing on this page has to be taken on trust. Each page states the vault path it came from. The vault file stays the source of record.
GitHub, Jira, the monorepo, the gates, the roles, the order of work. Verified 2026-08-01, re-checked today.
The complexity audit ruling: 5 keeps, 7 cuts, 4 defers. Read before any other sprint document.
How an idea becomes a shipped feature. The seven stages and the definition of ready.
Monorepo layout, the GitHub and Jira process, CI gates, owners and first branches.
Welcome, process introduction, roles, working agreements.
24 candidates, 6 run modes, the sprint-1 cut.
Component inventory and migration destinations, adjudicated.
Function-to-platform table and the GitHub system-of-record rules.
The five files staged on disk that populate the repository the day it exists.
The authority document. What runs, what is prototype, what is specified and unbuilt.
What each component does and how the three-layer knowledge model composes.
The operating loop and every agent in it, each marked run-today, semi-auto, or prototype.
Orientation and reading order for the pack that replaced the agent's wrong model.
Discard the six repositories by name, then report what you now hold.
Real components only, the tense rule, the required verification section.
What DKR means, how Scout finds gaps, and the writing rules that gate all of it.
| Site | Link | Note |
|---|---|---|
| GitHub organization | github.com/shuriq-lab | Zero repositories as of 2026-08-04 |
| Sprint 1 team document | Google Doc | The five drafts compiled and team-editable |
| Report Studio context, for Alex | Google Doc | Full-context brief |
| Report Studio, running | shuriq-report-studio.pages.dev | Production |
| Content series, running | content-series-b6l.pages.dev | 13 registered series |
What has to be true by the time we hang up, and what happens in the 48 hours after.
monorepo-seed/.route-gate.py opened as the first pull request, with the selftest wired into Actions.Where work items live (D-4), whether a quality assurance function exists, and when Jira mounts. All three are on the list above and all three can survive another week without blocking a build.