AI-DLC Context Engineering Playbook
Every Bolt is only as good as the context the AI was given. Is anyone actually building it?
AI-DLC's velocity claims rest on one assumption: that the AI plans and proposes correctly on the first pass. Whether it does is determined almost entirely by the quality of the context it receives — and under traditional methodologies, that knowledge lived informally in people's heads and was never a named responsibility. This playbook treats context engineering as a first-class AI-DLC discipline: five knowledge artifact types, file design patterns, a brownfield extraction method, four anti-patterns, a governance model, a full worked example, and five ready-to-use templates an engineering lead can put to use within the hour.
Delivered as a typeset PDF playbook. Download immediately after purchase.
■ Instant download · ■ No software required · ■ Read on any device
The AI-DLC Context Engineering Playbook is a practitioner guide defining context engineering as a named AI-DLC discipline. It sets out five knowledge artifact types — Design, Architecture, Product, Conventions, and Domain — a recommended file structure (YAML frontmatter plus structured markdown), a brownfield extraction method for teams without existing artifacts, four anti-patterns that quietly degrade AI output (context bloat, stale context, sprawl, and missing domain context), a governance model with named ownership, a full worked example across Inception, Construction, and Operations, and five ready-to-use templates. It is complemented by a free companion agent skill, the AI-DLC Context Pack Builder, which assembles a single Bolt's context pack with an automated pass/fail quality gate.
Poor context doesn’t announce itself. It shows up disguised as other problems.
Ritual cadence and governance gate design can be adapted from disciplines an organisation already knows. Context engineering cannot — the knowledge an AI now needs used to live informally in people's heads, transmitted through conversation and time served on the team. Under AI-DLC, that knowledge has to become a written, owned, and maintained artifact, or the AI is guessing.
The practical test: if a competent engineer joining the team tomorrow could not answer a question from the written context artifacts alone, neither can the AI — and the AI hits that gap on every single Bolt, not once per new joiner.
Five artifact types. One density principle. A named owner for each.
The playbook organises persistent context into five artifact types, each with a different natural owner, update rhythm, and failure mode when neglected. Collapsing them into a single “project context” document is the fastest route to the bloat and sprawl anti-patterns covered in Section 6.
The Five Knowledge Artifact Types · Section 2The semantics-per-token principle. Context quality is not volume, it is information density — the amount of decision-relevant meaning carried per token the AI must process. When in doubt about whether something belongs in a context artifact, the question is not “is this true?” but “would this change what the AI proposes?” If it would not, it is dilution.
- Context Steward
- A team-level role, often rotating or shared, accountable for the health of the context system as a whole: coverage across all five types, freshness discipline, and resolution of contradictions between artifacts. Works alongside the AI-DLC Coach.
- Freshness Window
- The period an artifact may go unreviewed before it is treated as suspect, declared in its own frontmatter and set by volatility rather than uniformly — the same window the Governance & Metrics Dashboard's Context Currency indicator tracks.
Every section, explained.
One PDF playbook, 11 sections, structured to be read once in full and then used as a standing reference.
Context Engineering Playbook vs. the rest of the AI-DLC line.
This playbook defines one specific discipline in depth. It is distinct from the vocabulary reference, the operational tracker, and the periodic maturity diagnostic — each answers a different question.
| Instrument | Question It Answers | Format |
|---|---|---|
| AI-DLC Context Pack Builder (Skill) | What's the minimum sufficient context for this one Bolt? | Free agent skill |
| AI-DLC Dictionary of Terms | What does this term mean? | Free web reference |
| Context Engineering Playbook | How do we build and govern the context AI-DLC depends on, across every Bolt? | PDF playbook |
| AI-DLC Governance & Metrics Dashboard | Is anything wrong right now, including context currency? | Excel workbook |
| AI-DLC Maturity Model & Benchmarking | How mature is our context management overall? | Excel workbook (not yet published) |
Not every Bolt needs this playbook’s full depth. Some just need one page, done right.
The AI-DLC Context Pack Builder is a free agent skill Viksya publishes alongside this playbook. It walks you through assembling a single Context Pack for one Bolt or Unit of Work — Intent, scope boundary, architecture touchpoints, contracts and constraints, definition of done, and references — then audits it in both directions against a pass/fail quality gate and won't hand it over until it passes. It is deliberately thin: the assembly procedure and the quality gate, nothing more.
Free, instant, no email required: plain markdown, works in Claude Code or any agent harness that reads a system prompt, and free to keep and share with attribution intact. (Placeholder link — page not yet published.)
Get the Free Skill →For whoever has to make the AI's first proposal the right one.
Structure the Architecture and Conventions artifacts that determine whether AI-proposed code fits the system.
Own the cross-cutting health of the context system — coverage, freshness, and resolving contradictions between artifacts.
Build the Domain artifact that keeps AI output commercially and operationally correct, not just technically correct.
Maintain the Product artifact that anchors every Unit of Work to business intent and explicit out-of-scope decisions.
Follow the structured extraction method in Section 5 instead of assuming the AI can infer years of undocumented decisions.
Feeds directly into: the AI-DLC Governance & Metrics Dashboard, whose Context/Knowledge Artifact Currency indicator tracks the same freshness window this playbook defines.
View Governance & Metrics Dashboard →Assigns ownership via: the AI-DLC RACI & Governance Gate Design Template, where artifact owners and the Context Steward role are formally recorded. (Placeholder link — product not yet published.)
View RACI & Gate Design Template →Companion instrument: the AI-DLC Role Transition & Org Design Guide covers who the Context Steward role actually is, and where it sits organisationally.
View Role Transition & Org Design Guide →New to the vocabulary? Intent, Unit of Work, Bolt, Mob Elaboration, Mob Construction, and human validation gates are all defined terms this playbook uses throughout.
View AI-DLC Dictionary →What makes this a practitioner playbook, not a documentation policy.
What this playbook is not.
This is a practitioner instruction manual for one AI-DLC discipline. It is not a scored diagnostic, a governance-assignment tool, a vocabulary reference in its own right, or the free single-Bolt skill.
What you need to use it.
Questions buyers ask before their first artifact.
We're greenfield. Is the brownfield extraction section still relevant?
Read it anyway. The four-step approach — static analysis, historical mining, interviews, validation — is written for inherited systems, but the underlying discipline of deliberately building context rather than assuming it applies from day one on a greenfield project too.
What's the difference between this playbook and the free AI-DLC Context Pack Builder skill?
The free skill assembles one Bolt's Context Pack — Intent, scope boundary, architecture touchpoints, contracts and constraints, definition of done, and references — and audits it against a pass/fail quality gate before you start. It is deliberately thin: procedure and gate, nothing more. This playbook is the standing, org-level counterpart: the five artifact types as living team artifacts, file design patterns, brownfield extraction, the full anti-pattern catalogue, and a governance model that survives across every Bolt, not just the one in front of you. Teams who hit a hard case with the skill — multi-team context, legacy systems, conflicting constraints — are exactly who this playbook is for.
How long before a team can produce its first context artifact?
The playbook is written so an engineering lead can structure a team's first artifact within the hour: Section 2 defines what to build, Section 4 defines how to structure it, and the Templates Appendix provides copy-paste starting points.
Do we need all five artifact types from day one?
Coverage should match where Bolts are currently running — the maturity checklist in Section 9 frames this as “all five types exist for every scope where Bolts are running,” not every scope in the organisation. Expand coverage as delivery expands.
How does this relate to the Governance & Metrics Dashboard we already use?
Directly. The Context/Knowledge Artifact Currency indicator in that dashboard tracks the same freshness window this playbook defines in artifact frontmatter (Section 4.4) — a declining currency reading there is the earliest observable signal of the stale-context anti-pattern this playbook diagnoses in Section 6.
Is this only for engineering roles?
No. The Domain and Product artifacts are explicitly designed to be owned by Business Analysts, domain SMEs, and Product Owners — the playbook treats missing domain context as one of the most common and most damaging anti-patterns precisely because it is often left to engineers by default.
What format does it come in, and can I use it internally?
A typeset PDF, delivered as an instant download. It is yours to keep and use for internal team reference and onboarding; it is not licensed for resale or redistribution as a standalone product.
Give the AI something real to reason from.
Structure your first context artifact this week, using the templates and worked example inside.
■ Instant download · ■ PDF playbook · ■ Yours to adapt for internal use