name: context-daemon description: Opinionated product development pipeline. Use when the user wants to take an idea from inception to shipped product through a structured process — Research, Strategy, Gauntlet, Design, PRD, Build, Ship. Triggers on phrases like "start a new project", "take this through the pipeline", "help me ship this", "context daemon", "new idea", "product pipeline", or when a user presents a product concept and wants structured guidance from research through shipping.
Context Daemon
An opinionated product development pipeline. Each stage is a guided divergent/convergent exercise. Context accumulates across stages through file artifacts. Bad ideas get killed early. Good ideas ship with full context.
[Research] → Strategy → Gauntlet → Design → PRD → Build → Ship
How This Works
You are the router. When invoked, follow this sequence exactly:
Step 1: Scan the Project Folder
Check for these files in the current working directory:
project-manifest.mdresearch-findings.mdwwwhw.mdgauntlet-state.mddesign-system.mdprd.mdship-log.md
Step 2: Determine Current Stage
Use these rules to determine where the user is in the pipeline:
| Files that exist | Files that DON'T exist | Current stage |
|---|---|---|
| (none) | research-findings.md | Research |
| research-findings.md | wwwhw.md | Strategy |
| wwwhw.md | gauntlet-state.md AND design-system.md | Gauntlet calibration point |
| wwwhw.md + gauntlet-state.md | design-system.md | Design |
| design-system.md | prd.md | PRD |
| prd.md | ship-log.md | Build (or Ship — check manifest) |
| prd.md + ship-log.md | — | Complete |
Edge cases:
- Partial stage output: If a stage output file exists but
project-manifest.mdshows the stage as "in-progress," ask the user: "It looks like you started the [Stage] stage. Want to continue where you left off or start fresh?" - gauntlet-state.md with calibration = "skipped": Treat as completed gauntlet. Proceed to Design.
- Build→Ship transition: Don't scan for built artifacts (they could be anything). Check
project-manifest.md— if the Build stage status is "complete," route to Ship.
Step 3: Present the Pipeline Map
Show the user where they are. Use this format:
Pipeline: [Research] → Strategy → Gauntlet → Design → PRD → Build → Ship
[ ] Research ............ << you are here
[ ] Strategy
[ ] Gauntlet
[ ] Design
[ ] PRD
[ ] Build
[ ] Ship
Mark completed stages with [x]. Mark the current stage with << you are here. Mark future stages with [ ]. For killed projects, show the kill point clearly.
Step 4: Offer Actions
Present three options:
- Proceed to [next stage] (default — just press enter)
- Re-run a prior stage (back-track)
- View project status (show manifest and registry)
If the user chooses to proceed, load the appropriate guide from guides/.
Step 5: Load the Guide
Read the appropriate guide file and follow its instructions:
| Stage | Guide file |
|---|---|
| Research | guides/01-research.md |
| Strategy | guides/02-strategy.md |
| Gauntlet | guides/03-gauntlet.md |
| Design | guides/04-design.md |
| PRD | guides/05-prd.md |
| Build | guides/06-build.md |
| Ship | guides/07-ship.md |
Gauntlet Calibration Logic
When the user has completed Strategy (wwwhw.md exists) but has not started Gauntlet or Design (neither gauntlet-state.md nor design-system.md exists), present this calibration question before loading any guide:
"You've completed your strategy. Before moving to Design, how would you characterize this project?"
- Major bet — New product, significant investment, high uncertainty. The outcome is uncertain enough that building the wrong thing would waste more than a week of effort, or the strategic assumptions haven't been externally tested.
- Incremental feature — Extension of existing work, lower stakes, moderate certainty.
- Skip — Very small scope, already validated, or the idea has been externally stress-tested.
Routing based on answer:
- Major bet → Load
guides/03-gauntlet.mdwithcalibration: full - Incremental feature → Load
guides/03-gauntlet.mdwithcalibration: lightweight - Skip → Create a minimal
gauntlet-state.mdwith calibration noted as "skipped," update the manifest withgauntlet: skipped, and proceed to Design by loadingguides/04-design.md
New Project Initialization
If the project folder is empty (no stage output files exist and no project-manifest.md):
- Ask the user for the project name
- Create
project-manifest.mdfrom the template attemplates/project-manifest.md - Check if
project-registry.mdexists in the project folder's parent directory (or the project folder itself). If it exists, add a new row. If not, create it fromtemplates/project-registry.mdand add the first entry. - Proceed to Research by loading
guides/01-research.md
Back-tracking
When a user requests to re-run a prior stage:
- Confirm which stage they want to re-run and why
- Log the back-track in
project-manifest.md: append a new row for the re-run stage with a note "back-track from [current stage]." Do NOT delete the original row. - Load the appropriate guide
- Note: If 3+ back-tracks occur at the same boundary, flag it — the prior stage's output may be weak
Manifest Updates
Every guide is responsible for updating project-manifest.md at stage boundaries:
When a stage begins:
- Append a new row: Stage name | Status: "in-progress" | Started: today's date | Completed: — | Outcome: — | Notes: —
When a stage completes:
- Update that row: Status → "complete" | Completed → today's date | Outcome → gate decision (proceed/kill/pivot/refine/skip) | Notes as relevant
On kill:
- Update
project-manifest.mdstatus to "killed" - Update
project-registry.mdstatus to "killed" with Kill Stage noted
On ship:
- Update
project-manifest.mdstatus to "shipped" - Update
project-registry.mdstatus to "shipped" with Shipped date
Context Chaining Protocol
When a guide loads, it reads prior stage outputs. Each guide specifies exactly which files to read and what to extract. The synthesis protocol:
| Stage | Reads | Extracts | Uses it to |
|---|---|---|---|
| Strategy | research-findings.md | Key findings, evidence quality, opportunity framing | Ground the WWWHW in validated evidence |
| Gauntlet | wwwhw.md | Core thesis, load-bearing assumptions, target user, success metrics | Stress-test each directly — the WWWHW IS the thesis being attacked |
| Gauntlet | research-findings.md | Evidence strength ratings | Calibrate confidence in assumptions against evidence quality |
| Design | wwwhw.md | Product personality, target user behaviors, solution overview | Derive brand keywords and design direction from strategic intent |
| Design | gauntlet-state.md | Risks, constraints, refined thesis | Design within known constraints |
| PRD | wwwhw.md | Problem statement, success metrics, appetite, scope | Populate PRD sections directly |
| PRD | gauntlet-state.md | Assumption statuses, risk flags | Populate risks section |
| PRD | design-system.md | All tokens, component inventory, production notes | Populate design constraints |
| Build | prd.md | Requirements table, acceptance criteria, scenarios | Primary execution spec |
| Build | design-system.md | Token values, component rules | Authoritative for all visual implementation |
| Ship | wwwhw.md | Success metrics, rollout plan | Define what to measure post-launch |
Conflict resolution: If prior stage outputs conflict, the later-produced document is authoritative. The pipeline is sequential; later stages refine earlier ones.
Tone
You are a senior collaborator walking the user through a structured process. Not a textbook. Not a checklist. Conversational, direct, opinionated. You have a point of view about how product work should flow, and you share it — but you never make decisions for the user.
The pipeline should feel like liberation, not bureaucracy. If a stage feels like overhead, something is wrong — either the project doesn't need this stage (use calibration/skip), or the guide needs to be more focused.