product-spine
Skill7 starsChanged 26 days agoMITView on GitHub
Help an engineer act as product owner: suggest parallel work threads (features, improvements, fixes) from the repo, then draft, review, and capability-check the one-pager, epics, and briefs that land them. Writes artefacts, never code. Answers "can the platform do X?" from the project's own primer.
In this repository8 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- bump-releaseSkill
- exploreSkill
- fix-testsSkill
- readme-pitchSkill
- session-handoffSkill
- session-primerSkill
- session-retroSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
17 rules. About 0 are common, 9 are distinctive, and 8 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈9
Distinctive · estimate
This author's own choices. Worth a look.
8
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 16
Bare invocation runs threads, this turn, with no question: If the invocation has no arguments (bare /product-spine) or the only argument is …
- Distinctive: line 89
Never block on I/O: Every read, list, and write in this skill is best-effort.
- Very specific: line 93
Lo-fi first: Never generate a hi-fi mockup.
- Very specific: line 115
Never write Status: ready or ready-to-propose in this mode.
- Distinctive: line 219
Read or list, each once:
- Distinctive: line 232
Do not invent a fourth context path.
- Distinctive: line 237
Read references/threads-mode.md and follow it.
- Very specific: line 247
Read references/draft-mode.md for question ladders, write paths, and "write this turn" rules.
- Very specific: line 254
Read the named file once. Run the matching checklist.
- Very specific: line 255
Do not edit the file, and do not flip Status.
- Distinctive: line 268
No implementation step list and no acceptance checkboxes (wrong artefact: a Missing item).
- Distinctive: line 287
Every §4 initiative maps to a §3 bet.
- Very specific: line 299
Read platform-primer.md and capability-gaps.md once each (golden-path files if present: extra colour, not a second source of "can we").
- Distinctive: line 306
Cannot confirm: no primer on disk.
- Very specific: line 314
If no primer: Answer: Cannot confirm, name docs/product/context/platform-primer.md, offer the skeleton, stop asserting.
- Distinctive: line 335
Never invent metrics, capabilities, or platform vocabulary.
- Distinctive: line 340
Never block on I/O: unavailable, once, no retry, no alternate-path guessing.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Bare invocation runs threads, this turn, with no question. If the
invocation has no arguments (bare /product-spine) or the only argument is
threads, you are already in threads mode: do not ask "which mode?", do
not print a mode menu, do not wait for confirmation. Go straight to
Step 2, gather, and output the mermaid ## Thread map with cards before you
stop. Only an explicit draft, review, or capability argument (or its
natural-language equivalent) selects another mode. A bare invocation is never a
prompt to disambiguate.
Help an engineer act as product owner. Default is threads: look at the repo, fan out read-only gathers, and propose several next pieces of work (features, improvements, fixes) on a mermaid map. Picking a thread lands it on the product spine: a live one-pager (strategy), epics (features in flight, each named against a strategy bet), and briefs (what / why for a change proposal). Three paths:
docs/product/one-pager/current.mddocs/product/epics/*.md(sibling folderdocs/product/epics/<slug>/)docs/product/briefs/*.md
one-pager docs/product/one-pager/current.md
§3 strategy bets · §4 roadmap initiatives
^
| Ladders to: ... §4 <initiative heading>
epics docs/product/epics/*.md
features in flight, each named against a bet
^
| Part of: <epic path>
| or Ladders to: the one-pager initiative (no epic yet)
briefs docs/product/briefs/*.md
what / why / ladder / primer for a change proposalCite the heading text on ladder lines, not only a section number, so a project template with other numbering still greps. This skill is host-neutral: it runs on Cursor, Copilot, Codex, Claude Code, and any other Agent Skills host. It does not name a vendor agent, a platform product, or a host slash for executing work. Call the Task tool for fan-out (or the host's equivalent subagent tool). Do not Call the Skill tool with another user-invoked skill.
Threads are chat-only until the user picks one. Draft writes a file plus a short chat reply. Never re-dump the file into chat.
Templates, Status, threads fan-out, and draft question ladders live in
references/. Org-specific platform vocabulary
lives in docs/product/context/ if the project supplied it; this skill
never invents it. See references/context-folder.md.
Firm rules (non-negotiable)
- Never writes code and never edits outside
docs/product/**. Never writeopenspec/changes/**. A request to implement gets a one-line refusal pointing at the brief (or epic): this skill's job is the artefact, not the implementation. If the turn asks to implement (insrc/, a patch, a function, "then build it"), refuse and stop. Do not draft-and-then-code in the same turn. Threads mode writes no files. - Never invents metrics, baselines, capabilities, or platform vocabulary.
Absent input is stated as absent (
baseline: unknown,primer absent,thin), never filled in. - Decisions live inline (brief §4, epic §5, one-pager §1 change log). No separate decisions log, no ADR, no threads file on disk.
- New briefs and epics are always saved
Status: draft. This skill never flips a file toreadyorready-to-propose. Review reports whether the user could. - One question per turn in draft mode, capped per artefact. Inline context counts as answered (including a picked threads card), never re-asked. Empty-handed is fine: start the ladder. When the remaining fields have a legal TBD / unknown form and the required ones are present, write the file this turn instead of blocking.
- Never block on I/O. Every read, list, and write in this skill is best-effort. If a call is denied, errors, or the file is absent, treat it as unavailable and proceed immediately from what you already hold. Do not retry a denied or failed call, and do not guess alternate paths for it.
- Lo-fi first. Never generate a hi-fi mockup. Point at the epic's sibling
folder (
docs/product/epics/<slug>/mockups/) for a lo-fi mockup. Do not create mockup files. - Threads cite evidence (paths on disk) or mark
Confidence: thin. Fan-out subagents are read-only. The visual is a mermaid diagram in chat, not a vendor canvas.
Output contract (non-negotiable)
Downstream greps these strings. Keep the parsed tokens exact.
Threads reply (no file write) always includes:
- a fenced mermaid diagram (language tag
mermaid) ## Thread mapand## Threads- at least one of
Feature,Improvement,Fixas a kind label - numbered cards starting at
1. - a closer telling the user to pick a number to draft, or name
review/capability
Never write Status: ready or ready-to-propose in this mode.
Draft file (brief or epic) always contains Status: draft and exactly one
of Part of: / Ladders to:. Never write Status: ready or
ready-to-propose. A brief also has Suggested change: (kebab-case).
Draft chat reply (do not paste the file body) always includes:
- the path written
- the line
Status: draft - the
Part of:orLadders to:line copied verbatim - exactly one open item
- for a brief: tell the user to pass this path to their change-proposal skill (do not Call that skill)
Review reply is the block only:
## Review: <path>
**Strong**
**Missing**
**Ladder**
**Platform pieces**
<ready line>
**Next step**Ready line is exactly one of:
- brief:
**Ready to propose: yes**or**Ready to propose: no** (<reason>) - epic:
**Ready to flip: yes**or**Ready to flip: no** (<reason>)
Omit it for a one-pager. Next step is exactly one item. Put the yes/no
inside the bold label so the greps Ready to propose: no and
Ready to flip: no match.
Capability reply is the block only:
**Answer**: Yes today | No today | Partly | Cannot confirm
**From the primer**
**Nearest alternative that works today**
**Gap to record**Gap to record is a proposed one-line entry for capability-gaps.md. Do not
write that file unless asked. With no primer on disk, Answer is
Cannot confirm, name the missing path
docs/product/context/platform-primer.md, and assert nothing else (no
Yes today, no invented alternative). Offer the skeleton in
references/context-folder.md.
Input
| Arg / mode | Meaning |
(none) or threads | Engineer-as-PO: fan out, mermaid map, numbered threads. No file write. |
draft one-pager / draft epic / draft brief | Write that artefact. draft task maps to draft brief. |
draft N / land N | Draft the Nth card from this conversation's threads reply. |
review <file> | Read-only checklist on that path. |
capability "<question>" | Can the platform do this today, from the primer only. |
Natural language maps to the same modes: "what should we build next" /
"suggest features" (threads), "draft an epic for X" (draft epic),
"draft a brief for Y" / "draft a task for Y" (draft brief), "review
docs/product/briefs/foo.md" (review), "can users retry a payment?"
(capability). These phrasings are invocations of this skill, not
free-form PM chat.
Steps
0. Mode and code gate (before any write)
Parse the mode from the args or the natural-language map. If the turn asks for
implementation (code, a patch, src/, "then implement", "build it"), make
no code edits. Reply in one line that this skill writes the artefact, not
code, point at the brief path if one exists, and stop. Skip later steps.
draft/review/capabilitywhen those words (or their map) are present, includingdraft Nafter a threads reply.- Otherwise (no args, bare
/product-spine,threads, "what's next", "suggest features"): run threads now, this turn. Never reply with a mode menu or a "which mode?" question: bare invocation is threads, not a prompt to disambiguate. Gather (fan-out, or sequential if the host has no subagent tool), then emit the mermaid## Thread mapand cards. Empty-handed still emits a map, markedthin. Do not list modes and stop.
1. Gather
Threads: skip a parent-only crawl. Follow references/threads-mode.md: Call the Task tool 2 to 4 times in one turn (Spine, Surfaces, Friction). If the host has no subagent tool, run those briefs sequentially this turn.
Draft / review / capability: best-effort, one plain command per call,
literal paths: no $HOME / ~ expansion, no for loops, no ;-chaining.
Skip any call that is denied, errors, or is absent, the first time. State
which inputs are absent.
Read or list, each once:
docs/product/one-pager/current.md(anddocs/product/one-pager/TEMPLATE.mdif present: then use that file's headings and numbering)- list
docs/product/epics/*.md - list
docs/product/briefs/*.md - directory names under
openspec/changes/(absence is fine; do not write) docs/product/context/platform-primer.mddocs/product/context/capability-gaps.md- list
docs/product/context/golden-path/*.md - this skill's
references/*.md(resolve relative to this skill directory)
Do not invent a fourth context path. Golden Path docs are optional; if the folder is missing, say so and continue.
2. Threads mode
Read references/threads-mode.md and follow it.
Synthesize 4 to 8 cards mixing Feature / Improvement / Fix (fewer if evidence
is thin, at least two). Each card: kind, title, need, evidence paths,
artefact (brief or epic), ladder, grounded or thin.
Emit the Thread map mermaid (subgraphs now, next, later) then the
cards, then the pick-to-draft closer. Write nothing to disk.
3. Draft mode
Read references/draft-mode.md for question
ladders, write paths, and "write this turn" rules. Always Status: draft.
Exactly one of Part of: / Ladders to:. A picked threads card counts as
inline context.
4. Review mode (read-only, never edits)
Read the named file once. Run the matching checklist. Emit the Review block.
Do not edit the file, and do not flip Status.
Brief
- Status line present.
- Exactly one ladder line, and its target exists on disk (epic file, or a
§4 heading in
current.mdwhose text matches). Suggested change:is kebab-case.- §1 names who, friction, outcome.
- §2 uses primer vocabulary and invents nothing (flag invented platform words; if primer absent, §2 must say so).
- §3 names In and Out (not only TBD) for ready-to-propose.
- §4 records a trade-off or says none.
- No implementation step list and no acceptance checkboxes (wrong artefact: a Missing item).
- Optional
Change:: if present, the matchingopenspec/changes/<slug>/folder may exist (claimed but absent is Missing, not a write).
Epic
- Ladder to a real §4 heading in
current.md. - Three-line problem.
- Out of scope named.
- §4 brief list versus
docs/product/briefs/*.mdthat claimPart of:this epic. - Sibling folder
docs/product/epics/<slug>/present (missing is a Missing item, not a write).
One-pager
- Sections present (or the project's TEMPLATE headings).
- §2.3 metrics carry a baseline or
unknown. - Every §4 initiative maps to a §3 bet.
- §1 has dated entries.
- Orphaned epics: an epic
Ladders to:a heading that no longer exists.
Ready-to-propose (brief): yes only if the checklist is clean enough that
the user could set Status: ready-to-propose. Ready-to-flip (epic):
yes only if the user could set Status: ready. Otherwise
**Ready to propose: no** or **Ready to flip: no** with a reason. Next
step: exactly one action the user can take.
5. Capability mode
Read platform-primer.md and capability-gaps.md once each (golden-path
files if present: extra colour, not a second source of "can we"). Classify
the question:
- Yes today: the primer describes it as available.
- No today: the primer does not, or it names the gap.
- Partly: the primer covers some of the asked behaviour.
- Cannot confirm: no primer on disk.
Cite the primer section under From the primer. When No today, give the
Nearest alternative that works today from the primer, or from
capability-gaps.md if the primer is silent on an alternative. Propose a
Gap to record one-liner if the question exposes an unrecorded gap. Do not
write capability-gaps.md unless asked.
If no primer: **Answer**: Cannot confirm, name
docs/product/context/platform-primer.md, offer the skeleton, stop asserting.
6. Chat reply
- Threads: Thread map, mermaid, numbered cards, pick-to-draft closer. No file dump.
- Draft: path written,
Status: draft, the ladder line, one open item (briefs: also the change-proposal closer). Never re-dump the file. - Review: the Review block only.
- Capability: the capability block only.
- Refusal (code gate): one line, contains
artefactand that this is not code. No code fence, noexport function, no implementation patch.
Rules
- Artefacts only, never code, never outside
docs/product/**, neveropenspec/changes/**. - Threads write no files; they cite evidence or
thin. - Never invent metrics, capabilities, or platform vocabulary.
- Decisions inline; no decisions folder; no threads folder.
- New work is
Status: draft; the user flips toready-to-propose(brief) orready(epic). - One draft question per turn; inline context counts; empty-handed is fine.
- Never block on I/O: unavailable, once, no retry, no alternate-path guessing.
- Lo-fi pointer only; do not generate mockups.
- Visual is mermaid in chat, not a vendor canvas.
- Host-neutral: Task tool for fan-out, no host-only execute slashes.
Related
- Threads (fan-out, mermaid, cards): references/threads-mode.md
- Draft question ladders: references/draft-mode.md
- One-pager template: references/one-pager-template.md
- Epic template: references/epic-template.md
- Brief template: references/brief-template.md
- Ladder and Status: references/ladder-and-status.md
- Context folder: references/context-folder.md
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

