with one click
agenthub-docs-spec
Canonical feature spec writing for stable contracts.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Menu
Canonical feature spec writing for stable contracts.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
| name | agenthub-docs-spec |
| description | Canonical feature spec writing for stable contracts. |
Use this skill when writing or reviewing docs/features/*.md files for stable runtime, UI, or API contracts instead of rollout history.
Keep feature specs as the stable source of truth for:
Feature specs are not date-stamped rollout notes and should not read like a changelog.
docs/features/docs/features/docs/journal/docs/todo.mdEach active feature spec should include:
ProblemScopeNon-GoalsArchitectureContractsValidation MatrixOperational NotesOpen RisksSource JournalsIf a section is intentionally short, keep it short, but do not silently drop it.
A spec should answer:
Avoid low-signal content such as:
Do not let a spec silently expand into adjacent product surfaces.
Examples:
When there is a transition period, distinguish:
Example shape:
/workspace/nodes/:node_id?lens=nodes&node=...Do not leave both paths sounding equally primary unless that is truly intended.
The Validation Matrix should describe the focused checks needed to trust the spec:
Every non-trivial spec should link the journals that established or evolved it.
This lets the spec stay compact while journals retain the chronology.
When multiple journals describe the same area:
docs/todo.md follow-ups at the canonical spec, not at drifting scratch contextCreate or update a feature spec when:
Do not use a spec for:
Check these:
docs/todo.md, not only in proseUse when a Team inbox, channel, thread, or human-visible message must be routed into a direct reply, thread reply, mailbox update, task note, or canonical Team task.
Use before editing Team system prompts, runtime prompt tails, prompt-linked skills, or prompt tests to keep prompts bounded, tool-neutral, and pointer-first.
ACP rendering fixes for Team web surfaces.
Backend-first diagnosis for stuck AgentHub agents.
Journal writing for rollout notes and validation evidence.
Focused AgentHub test planning for behavior changes.