Shared reference library for structured decision-making across the skill pipeline. These references codify the common vocabulary, formats, and protocols used by skills that work with product strategy, project decomposition, story sharpening, feature specification, and any other skill that produces durable artifacts.
This is not a workflow skill — it has no session arc, no phases, no interactive loop. It is a set of reference files that consuming skills load as needed.
When this skill gets loaded
If you loaded this skill directly (or it was loaded as a dependency), read the reference files relevant to your current task and adhere to them. The "Load when..." column in the table below tells you which files apply.
Available references
| Reference | What it provides | Load when... |
|---|---|---|
references/decision-taxonomy.md |
Resolution statuses (LOCKED/DIRECTED/DELEGATED/INVESTIGATING/DEFERRED/ASSUMED), assumption lifecycle, temporal non-goals (NEVER/NOT NOW/NOT UNLESS), confidence vocabulary, cascade analysis protocol, bounded reasoning, disagreement protocol | ...you are recording decisions, managing assumptions, deferring scope, or need to communicate decision latitude to downstream consumers |
references/artifact-discipline.md |
The four-layer model (conversation/primary artifact/evidence/changelog), dual-audience design (human + AI agent), evidence file conventions, write trigger protocol, cascade protocol, multi-session resume discipline | ...you are producing durable artifacts (specs, project docs, strategy docs, story seeds) that must survive context compression and be consumable across sessions |
references/problem-framing.md |
SCR format (Situation → Complication → Resolution), 5-probe stress test (demand reality, status quo, narrowest wedge, observation, future-fit), JTBD anchoring | ...you are framing a problem at any level — strategic bet, project scope, story definition, or feature spec |
references/disambiguation-protocol.md |
The 5-step disambiguation discipline: challenge what's stated, probe what's unstated, surface what's ambiguous, explore what's adjacent, verify what's assumed | ...you are in any interactive session where the output must be precise enough that the next consumer doesn't re-discover what could have been captured upstream |
references/challenge-posture.md |
Co-driver stance (opinions, push back, enforce rigor), anti-sycophancy rules (no agreement phrases, take positions with evidence), investigate-vs-judgment boundary, contamination awareness (consensus vs unique insight) | ...you are co-creating with a human in any interactive session — strategy, decomposition, sharpening, specification, analysis |
references/value-dimensions.md |
Why intersection reasoning matters more than individual dimensions, dimension sets by work level, the "percolation of why" problem, value connections (immediate/lateral/forward), the dimension-trace diagnostic | ...you are articulating why work matters, decomposing work into stories, ensuring value rationale carries through to downstream consumers, or verifying that planned work traces to genuine value |
references/extraction-protocol.md |
Three extraction probes (walk-through, tensions, negative space), "list without filtering" discipline, unified Items table schema + status lifecycle, load-bearing heuristic, carry-forward discipline, depth-level examples, vocabulary mapping to decision-taxonomy.md, evidence file types | ...you are systematically surfacing open questions, tracking decisions and assumptions in a unified Items table, or carrying forward tracked items between pipeline stages |
references/session-discipline.md |
Investigation escalation ladder, multi-answer parsing protocol, progress scorecard, interaction cadence (batch sizing, confidence-weighted presentation) | ...you are running an interactive session that involves investigating items, presenting batched decisions to users, and tracking progress across multiple interaction rounds |
How consuming skills reference these files (guidance for skill authors)
When writing a skill that depends on structured-thinking, use this pattern in your SKILL.md:
**Load:** Load /structured-thinking skill and read its reference file `references/decision-taxonomy.md`
The agent loads this SKILL.md (seeing the table above), then reads the specific reference file. This works because the agent resolves the path relative to this skill's base directory.
The "Extends:" convention: When a consuming skill has its own reference file that builds on a structured-thinking reference, that file uses an **Extends:** directive at the top:
**Extends:** Load `references/decision-taxonomy.md` from /structured-thinking for the shared
vocabulary. This reference adds skill-specific extensions on top.
This tells the agent: "read the structured-thinking reference first (the foundation), then continue reading this file (the skill-specific layer)." The consuming skill's SKILL.md should load /structured-thinking early in its workflow so the dependency is already satisfied when the agent encounters the Extends directive.