| name | roadmap-conventions |
| description | Encodes PromptJang's quarterly-roadmap conventions so agents reason consistently. Use when reviewing roadmap items, prioritization decisions, capacity envelopes, or forming agent positions on decision_required items. |
Roadmap Conventions Skill
Goal
Keep both agents (Planning + Stakeholder) reasoning from the same vocabulary
and thresholds when they debate the Q3 plan, so their disagreements are about
judgment, not about what words mean.
Conventions
Status taxonomy (the messy carry-over states)
completed — shipped, done.
in_progress — being worked now; spills into next quarter.
partially_done — some milestones hit, some not.
blocked — stuck on a dep (external team / vendor / design / infra).
not_started — planned, never began (lost to capacity).
cut — deprioritized / abandoned.
planning — proposed for the next quarter; not yet committed.
Decision types (what each decision_required item asks)
prioritize_vs_deprioritize — ship now vs defer.
unblock_vs_cut — invest in clearing the blocker vs drop the feature.
prioritize_vs_defer_partial — ship in full vs scope down.
auto_keep / auto_prioritize — both agents agree; no human decision needed.
Utilization thresholds (the Planning Agent's evidence)
[85%, 100%] — healthy band.
>100% — over-commitment; cite the Q1 122% peak as the cautionary precedent.
<85% — under-utilization (not currently a risk in this dataset).
Capacity envelope (the forcing function)
- Q3 initiative budget: 2,400h (after ~49% on-call / BAU).
- Demand: 2,660h (640h carry-over + 2,020h new proposals).
- Delta: +260h. At least one
decision_required item must be deprioritized,
cut, or deferred — the UI's Q3 Committed column refuses the full set.
Position-label vocabulary (what the agents emit)
prioritize | deprioritize | unblock | cut | defer_partial | auto_keep | auto_prioritize
Constraints
- Reasons MUST cite one of: incoming_state, feedback, or the capacity envelope.
Vague reasons ("it's important") are not acceptable.
- Never auto-commit a decision. Every
decision_required resolution ends at a
human Approve gate in the dashboard.
- Employee names +
(mock) suffixes must already be redacted before the agent
sees any text (handled by app.tools.redact_confidential upstream).