DKR Agency — Operations Guide
How to actually run the agents and reports that make up the DKR Agency today, and how they meet the SBPI cadence. This is the working manual; the conceptual frame is in DKR-SYNTHESIS-OUTLINE. Status of each tool is marked RUN-TODAY, SEMI-AUTO, or PROTOTYPE.
Draft — to be hardened during this week's sprints. Where a step is not yet wired, it says so.
0. The operating loop
Everything here serves one loop, run on a pulse:
NIGHTLY / WEEKLY PULSE
Scout → find what's missing (negative space)
Trailblazer→ map the environment + model intent (PGORC) [PROTOTYPE]
Social-Int → enrich the people in the picture [RUN-TODAY]
SBPI → score the competitive structure
Reports → daily / weekly / value-flow synthesis
DKR agent → link new notes, raise legibility, message the user [PROTOTYPE]
The first three gather and structure; SBPI scores; the reports synthesize; the DKR agent closes the loop by connecting everything and surfacing the next action. Each section below covers what it does, how to run it, what it outputs, and where it hands off.
1. Scout — gap-finding (RUN-TODAY)
Canonical spec: scout-agent (v1.1.0, approved). Onboarding: team-onboarding-scout.
What it does. Negative Space methodology — "structure by subtraction." Instead of "what do we know?" it asks "what are we not discussing?" across five question types:
- Absence — topics conspicuously missing given stated goals (coverage < 5%)
- Bridge — connections between clusters never explicitly made
- Decay — topics that dropped off but stay relevant (> 50% decline over 4+ weeks)
- Contradiction — unresolved tensions across conversations
- Horizon — adjacent fields we aren't monitoring
How to run it.
- On demand: /scout or /gaps against a topic, project, or client context.
- Scheduled: weekly internal + weekly external, monthly full (Monday 08:00 per spec config).
- It pulls graph topology (Knowledge Agent), recent synthesis (Intelligence Agent), and — for external/horizon — InfraNodus network analysis (betweenness centrality, community detection).
Output. A structured gap report: structural opportunities, research questions, content angles, priority scores (0-1), recommended actions. Stored under the project's gap_reports/.
Hands off to. Content Factory (content angles), Knowledge Agent (missing-bridge suggestions), Orchestrator (routing). Gaps scored > 0.7 can trigger a gap→visual-brief.
DKR Agency role. Scout is the Porter's Force Field + Horizon organ — it tells the DKR agent where the knowledge graph is thin so the nightly loop knows what to build toward.
2. Trailblazer — the agency harness (PROTOTYPE)
Source: Trailblazer Agent Design Ideas (design only — no canonical spec yet; this is sprint #1).
What it is meant to do. The harness that Scout, 3x3, and the DKR agent "hitch a ride" on while I do knowledge work. Tag-team orchestrator with MetaTotem (MetaTotem = conventional PM, persistent memory, its own Gmail/Drive, Obsidian + Bases expertise).
Designed capabilities. - Browser recording (Playwright / Claude-in-the-browser): records my actions while logged into any account, to model how I get things done so it can replicate the workflow. - Environmental state-space snapshot: captures the active and potential systems, markets, rules, processes, ledgers, and communication networks relevant to a piece of work. - PGORC model: a Polytopolis-style Preference / Goal / Outcome / Risk / Constraint model, visualized as bubbles with edges between current, sub-optimal, and optimal states (the wireframe design). - Cross-vault operation + matchmaking: works across multiple people's vaults, shared docs, Drive folders, Slack channels; nudges interactions and resource deployment toward higher shared outcomes.
How to run it. Not runnable yet. This week: write the canonical spec (frontmatter, version, approval), then a first prototype that does environmental-snapshot + PGORC on one real workflow.
DKR Agency role. Trailblazer is the connective tissue — the layer that lets the DKR agent operate across systems instead of inside one vault. The single highest-leverage gap.
3. Social-Intelligence agent — people enrichment (RUN-TODAY)
Source: MEMO-Social-Intelligence-Demo-CondoSales + the 8-hour build transcript. Production Python.
What it does. Enriches CRM leads (or any list of people) with verified social profiles and psychographic intelligence: LinkedIn / Instagram / GitHub / Facebook / X, detected interests, life stage, safe actions, and a 0-100 confidence score that gates how personally you can act on a finding.
How to run it. Two-phase hybrid:
- Phase 1 — URL discovery (collect_urls_only.py): adaptive search loop (4-6 strategies), location verification.
- Phase 2 — deep research (manus_automation.py / Kimi web agent): psychographic profiling, write-back to Google Sheets in real time.
- Resumable batches via run_slow_batch.py.
The trade-off (memorize this). Fast / Affordable / Accurate — pick two.
| Approach | 50 prospects | Use when |
|---|---|---|
| Kimi web agent (slow) — default | cents, ~1 day | continuous background enrichment |
| SerpAPI (fast) | ~$3 | urgent, pre-meeting |
| Manus (deep) | ~$50 | high-value single targets |
| Manual | 10-20 hrs | gold-standard sample / training |
Confidence gating: 80-100 → safe to personalize; 60-79 → use subtly, verify; 40-59 → professional only; 0-39 → generic templates.
DKR Agency role. This is how the DKR agent turns relationships into structured, scored intelligence — the soft/social-capital layer made legible. It feeds matchmaking.
4. The reports — daily / weekly / value-flow
4a. Daily report (SEMI-AUTO)
Source: RECIPE-REGISTRY → daily-digest recipe, launchd at 6 PM weekdays, plus the /daily-synthesis skill. Appends to the daily note's ## Context Inference. Health-check fires every 15 min and writes ## Alerts.
4b. Weekly report (SEMI-AUTO)
Source: system/schedules/Analysis-Schedule.base → weekly client digests (7 clients). Tool chain per job: generate_knowledge_graph → generate_content_gaps → generate_research_questions → develop_text_tool. Execution is currently manual from the Today's Queue; the queue logic exists, the auto-trigger does not.
4c. Value-flow mapping (RUN-TODAY, template exists)
Source: VALUE-FLOW-REPORT — the report I made for Limore. Method: ValueFlows REA ontology (McCarthy 1982) → InfraNodus graphs. Models four value dimensions (Client Billable / Investor Demonstration / Direct Revenue / Agency Acquisition), maps agents-resources-events, and surfaces structural gaps with actions.
How to produce one. Use the value-flow-ontology skill: extract agents/resources/events from the business material → build the REA graph in InfraNodus → run gap analysis → write the four-dimension report → create the Bases dashboards (*-Value-Flows.base, *-Stakeholders.base). The Limore report is the worked template.
DKR Agency role. Value-flow mapping is the CODIAK "improve the work" organ — it makes the business's own value structure a knowledge graph the DKR agent can reason over.
5. Where it all meets — the SBPI junction
The DKR operations connect to the SBPI engine at two scheduled moments. This is the integration the note calls out explicitly.
SBPI canonical: sbpi-methodology — 100-pt composite over 5 dimensions (Content Strength, Narrative Ownership, Distribution Power, Community Strength, Monetization). Brands stack-rank only within the same vertical.
| Moment | What runs | Wrapper | DKR handoff |
|---|---|---|---|
| Nightly insight | semantic-layer pipeline: Oxigraph RDF, prediction experiments, KG optimization | sbpi-nightly-insights.sh (live; last 2026-05-27) |
DKR nightly loop should read the night's deltas and link new notes to changed scores — not wired yet |
| Weekly report | stack-ranking synthesis across the vertical | sbpi-weekly-report.sh (live; ~2026-05-24) |
Scout reads the weekly stack rank → runs negative-space on the leader → feeds Content Factory |
The loop in one sentence: Scout finds the gap → Social-Intelligence enriches the people inside it → SBPI scores the competitive structure → value-flow maps how value moves → the reports synthesize → the DKR agent links it all to my goals and messages me the one thing to do next.
6. Cadence summary (what fires when)
| When | Job | Status |
|---|---|---|
| every 15 min | health-check | ✅ launchd |
| nightly | SBPI insight pipeline | ✅ live |
| nightly (4 AM) | DKR connection-making loop (dkr-nightly) |
✅ launchd |
| 6 PM weekdays | daily-digest | ✅ launchd |
| weekly | SBPI weekly stack rank | ✅ live |
| weekly | client digests (7) | ◐ manual from queue |
| weekly | Scout internal + external | ◐ on-demand today |
| Sun 2 AM | eval-runner | ✅ launchd |
| on Stop | content-map-updater | ✅ hook |
| per engagement | value-flow mapping | ✅ template |
7. Open items this guide will absorb (post-sprint)
- Trailblazer run-book (after the prototype lands).
- DKR agent onboarding script — the interrogation question bank + grammar.
- The nightly connection-making loop as a documented recipe in RECIPE-REGISTRY.
- Knowledge-cohort sharing + matchmaking run-book.
→ Build sequence: SPRINT-PLAN.