isvlasov Creating Briefs
A brief transforms an unstructured request into a bounded, verifiable specification that the executing agent can work from without clarification questions — and that the requester can read and say "yes, this is what I meant". Briefing is orchestrator-owned: it depends on the full conversation context, so write the brief yourself rather than delegating it.
Skip when the task is trivial, already well-specified, or purely exploratory with undefined outcomes.
Principles
- Outcomes, not solutions — define WHAT to achieve, let the executor determine HOW. Prescribe method only when a real constraint requires it.
- Context is never assumed — the executor hasn't read the conversation, previous work, or project history. If background matters, state it in the brief.
- Fit for purpose AND audience — name who will use the deliverable; it shapes requirements and tone.
Workflow
1. Clarify the request
Extract: core intent (the outcome behind the literal words), purpose (WHY it matters — without it the executor can do technically correct work that serves the wrong goal), scope boundaries, audience, and any assumptions you're making. If critical information is missing, ask targeted questions — don't guess.
2. Identify applicable standards
The 1–3 skills or guides that govern quality for this artefact type: writing-skills for skills, conducting-research for research, style guides for documents. Multiple standards can apply to one brief.
3. Extract requirements
The core step — it creates the shared quality expectations that the executor works to and the critic reviews against. From each standard, select the 5–12 requirements that apply to this task (not all of them), rephrase for clarity if needed, and structure as verifiable checkboxes grouped by source standard. Vague aspirations ("must be good quality") are not requirements.
4. Set constraints
3–6 scope limits: format, length, dependencies, explicit exclusions. More than that suggests the scope itself is unclear.
5. Define success criteria
2–4 measurable, outcome-focused statements the requester can verify ("QA can test all endpoints without questions", not "documentation is comprehensive"). The final criterion is always "all requirements from standards are met".
6. Add references and context
Any materials provided for the task — skills, previous work, examples — MUST appear in References; they are inputs, not nice-to-haves. In Context, summarise background the executor needs in 2–5 bullets: decisions already made, known risks. Need-to-know only — no speculation, no disguised requirements. Omit the section if there's nothing.
7. Validate
Before handing off: would the requester recognise this as their intent? Can the executor start without questions? Are requirements checkable, success criteria measurable by a third party, and all provided materials referenced? Fix gaps before use — a technically perfect brief that doesn't match the request is a failure.
Template
# Brief: [Task Name]
**Created:** YYYY-MM-DD
**Objective:** [One sentence outcome that's verifiable]
**Audience:** [Who will use the deliverable]
**Applicable Standards:** [Standards/skills that define quality]
## Why
[2–4 sentences: why this work matters, where it fits in the bigger picture,
what problem it solves]
## Requirements Extracted from Standards
**From [standard-name-1]:**
- [ ] Requirement 1 (specific, verifiable)
- [ ] Requirement 2 (specific, verifiable)
**From [standard-name-2]:**
- [ ] Requirement 3 (specific, verifiable)
## Constraints
- Format: [Specific requirement]
- Length: [Target or range]
- Dependencies: [What must exist]
- Out of scope: [Explicit exclusions]
## Success Criteria
The deliverable is complete when:
1. [Measurable outcome 1]
2. [Measurable outcome 2]
3. All requirements from standards are met
## References
**Skills:** [Path to skill files]
**Previous work:** [Path to research or related artefacts]
**Examples:** [Path to reference materials]
## Context
**Background decisions:** [Constraints from previous work]
**Known risks:** [Challenges to anticipate]
Typical length: 50–150 lines depending on task complexity.
Edge cases
- Conflicting standards — note the conflict in Context, propose a resolution based on artefact type and audience, ask the requester if unclear.
- No formal standards — derive checkboxes from best practice for the artefact type and label the section "Quality Expectations" to signal they're derived, not extracted.
- Prior research exists — include it in References and summarise the key findings in 2–4 Context bullets; extract any quality criteria the research surfaced into requirements.
- Iterating on an existing artefact (vN → vN+1) — reference the previous version and its review, add a "Changes from vN" section listing what the iteration addresses, keep still-applicable requirements, and add a success criterion about preserving what worked.
- Vague request — ask targeted questions; document confirmed assumptions in Context.