≈1.5k tokens on demand. Your agent reads this file only when SKILL.md points to it.
Cast Generation Workflows
Purpose: Define the CONJURE workflow, source detection order, synthesis rules, validation thresholds, and registration behavior.
Contents
- Source detection
- Input quality
- Extraction strategy
- Synthesis rules
- Validation
- Registration
- Auto-triggers
- Internal personas
Source Detection
When no explicit path is provided, detect sources in this order.
External / User Persona Sources
| Priority |
Source |
Use for |
Confidence contribution |
P0 |
.agents/personas/ |
Adopt or enrich existing personas |
— |
P1 |
README.md |
User types, product promise, usage context |
+0.10 |
P2 |
docs/**/*.md |
User guides, personas, workflows |
+0.15 |
P3 |
src/**/auth*, src/**/user* |
Roles, auth, user models |
+0.15 |
P4 |
src/**/checkout*, src/**/onboard* |
Key journeys, conversion friction |
+0.15 |
P5 |
tests/e2e/** |
User stories encoded in tests |
+0.10 |
P6 |
package.json, *.config.* |
Product metadata and flags |
+0.05 |
Internal Persona Sources
| Priority |
Source |
Use for |
P0 |
CODEOWNERS |
Team structure and responsibility |
P1 |
docs/CONTRIBUTING.md |
Development process |
P2 |
.vscode/**, .idea/**, .editorconfig |
Editor/workflow setup |
P3 |
docker-compose*.yml |
Local dev context |
P4 |
.github/workflows/* |
CI/CD patterns |
P5 |
Makefile, scripts/** |
Commands and automation |
P6 |
docs/runbook*, docs/onboarding* |
Operational and onboarding context |
Input Quality Assessment
| Factor |
High |
Medium |
Low |
| Explicitness |
User type stated directly |
Implied |
Guessed from filenames |
| Recency |
Updated within 30 days |
Updated within 90 days |
90+ days stale |
| Specificity |
Named user types with details |
General user mention |
No user mention |
| Breadth |
Multiple dimensions covered |
Partial coverage |
One dimension only |
Extraction Strategy
| Source type |
Extract |
| Documentation |
user types, goals, pain points, usage scenarios, language cues |
| Code |
roles, flows, permissions, friction points, feature flags |
| Tests |
user stories, expected flows, success criteria, fixtures |
| Field handoff |
quotes, goals, pain points, segment findings |
| Trace handoff |
behavior clusters, device mix, drop-off points |
| Voice handoff |
feedback themes, promoters/detractors, segment drift |
Extraction Output
Produce structured extraction output with:
service_contextuser_typesgoalspain_pointsbehaviorscontextsevidence
Synthesis Rules
- Generate at least
3 personas by default: P0, P1, P2.
- Assign an
echo_base_mapping to every persona.
- Record evidence for every non-trivial attribute.
- Mark unsupported attributes as
[inferred].
- Calculate confidence from source and validation contributions.
Priority Assignment
| Priority |
Criteria |
P0 |
Most common or strongest evidence |
P1 |
Second-most common or business-critical |
P2 |
Edge case, underserved segment, or high-friction segment |
P3+ |
Only if evidence strongly supports additional segments |
Detail Level Selection
| Condition |
Detail level |
| User explicitly requests a level |
Requested level |
4+ extended dimensions available |
Full |
2-3 extended dimensions available |
Standard |
type == internal |
Internal |
| Otherwise |
Minimal |
Validation
Echo Compatibility
| Check |
Rule |
Failure action |
| Frontmatter |
Echo-required fields present |
Add safe defaults |
Profile |
Includes core role/context fields |
Infer cautiously from evidence |
Quote |
Non-empty |
Compose from goals and frustrations |
Goals / Frustrations / Behaviors |
Keep 3 each |
Merge or split to reach 3 |
Emotion Triggers |
Cover the expected polarity spread |
Generate from context |
Echo Testing Focus |
At least 3 flows |
Derive from JTBD |
Source Analysis |
At least one source |
Block generation |
Consistency Checks
- Tech level must fit behaviors.
- Device emphasis must fit context scenarios.
- Goals and frustrations must not contradict each other.
- Echo mapping must match the dominant trait.
Confidence Thresholds
| Confidence |
Status |
Action |
>=0.60 |
active |
Ready for use |
0.40-0.59 |
draft |
Usable with enrichment warning |
<0.40 |
draft |
Warn and ask before relying on it |
Registration
File Operations
- Create
.agents/personas/{service-name}/ if missing.
- Write persona file as
{persona-name}.md in kebab-case.
- Update
.agents/personas/registry.yaml.
- Return a concise registration summary.
Naming Rules
- Service name: kebab-case.
- Persona name: kebab-case.
- Internal personas stay in the same service directory and are distinguished by
type: internal.
Registration Summary
Return:
- service
- personas generated
- detail level
- registry updated
- persona table with
priority / echo mapping / confidence / file
- analyzed sources
- recommended next command
Auto-Detection Triggers
Suggest Cast automatically when:
- Echo has no saved personas.
- Field completes persona creation.
- Trace delivers behavioral drift data.
- A new service is added to the project.
Internal Persona Rules
- Set
type: internal.
- Include
Internal Profile and Workflow Context.
- Map to an internal Echo base persona.
- Focus emotion triggers and testing focus on workflow and DX concerns.
Job Type Detection
| Signal |
Inferred type |
@frontend-team in CODEOWNERS |
Frontend Developer |
@backend-team in CODEOWNERS |
Backend Developer |
@infra-team in CODEOWNERS |
Infra Engineer |
@qa-team in CODEOWNERS |
QA Engineer |
| Storybook / Figma references |
UI Designer |
docs/runbook* |
Ops Manager |
admin/ directory |
Content Editor / CS |
docs/onboarding* |
New Engineer |
