Doccraft Session Wrap

doccraft — session wrap (propose docs artifacts)

When to use

• The user invokes this skill or asks to wrap this session /

propose docs artifacts / what should we capture / summarize this discussion into docs from the current conversation.

• Typical triggers: architecture or product discussion concluded (or

deferred with rationale), research synthesis, or reprioritisation — not every casual Q&A.

Default: propose artifacts only if necessary. Do not invent work to document.

NEVER write files unless the user explicitly asks to create or update them in the same turn. Propose drafts only.

Configuration

Read doccraft.yaml at invocation. The sessionWrap.capture: block controls which artifact categories are in scope for this project — skip rows entirely in the proposals table when their category is disabled.

Relevant keys:

• docsDir — root folder for all docs. Default: docs. All artifact

paths below are relative to {docsDir}/.

• sessionWrap.capture.research — default true.
• sessionWrap.capture.reference — default true.
• sessionWrap.capture.business — default false. Projects that track

strategy under {docsDir}/business/ set this to true; most projects leave it off.

If doccraft.yaml is missing, apply the defaults above. The ADR, Story, and Backlog / queue categories are always in scope — they are the load-bearing set.

Package routing

For monorepo projects (a Known package roots block appears above when packages are declared), artifact proposals route to the package whose body the conversation referenced most. Default to the project root when no single package dominates. The paths in the docs map below resolve relative to the chosen package's {{DOCS_DIR}}/ (or the project-root {{DOCS_DIR}}/ for single-root projects).

Docs map (where things go)

Authoritative overview, if the project has one: {{DOCS_DIR}}/README.md.

| Kind | Path | Use for | |------|------|--------| | ADR | {{DOCS_DIR}}/adr/NNN-slug.md | A decision to freeze (chosen option, explicit deferral, or rejected approach you do not want re-litigated). See doccraft-adr if installed. | | Research | {{DOCS_DIR}}/research/.md | Exploration / comparison / notes (tools, papers, datasets) without a single "we decided X" — synthesis for humans and future agents. | | Reference | {{DOCS_DIR}}/reference/.md | Long-form engineering notes (runbooks, eval harnesses, pricing) that are not framed as an ADR. If the project has a {{DOCS_DIR}}/reference/README.md, add a link there when adding a new file. | | Business | {{DOCS_DIR}}/business/.md | Business strategy — target audience, business model, competitive landscape, marketing, unit economics, legal, launch sequence. Only propose if the project already tracks business docs under {{DOCS_DIR}}/business/; prefer updating existing topics over creating new ones. | | Story | {{DOCS_DIR}}/stories/.md | Trackable scope with acceptance criteria and optional depends_on. Follow doccraft-story if installed. | | Backlog / queue | {{DOCS_DIR}}/backlog.md, {{DOCS_DIR}}/queue.md | P-tier rows, status column, ordered "pick next" — not every chat. If the thread defined new dependencies, point the user at doccraft-queue-audit (Agent mode applies mechanical fixes). |

Gate checklist (run mentally before proposing)

Answer each only from the visible thread (and optional repo files the user already opened). If the answer is "no" for all actionable rows, go to Exit.

1. ADR — Did the thread reach (or explicitly record) an **architecture /

process / stack** conclusion worth git history? (Includes "we will not do X".)

1. Research — Is there durable synthesis (comparisons, constraints,

citations) that is not a single sharp decision?

1. Reference — Is there operational or eval depth better as a living

doc than an ADR or a story?

1. Business — If the project tracks business strategy under

{{DOCS_DIR}}/business/: did the thread produce business insights (audience, pricing, competitive, marketing, legal, economics) that MUST update or extend those docs? Prefer updating existing docs over creating new ones. Skip this row entirely if the project has no {{DOCS_DIR}}/business/ tree.

1. Story — Is there a bounded deliverable with testable acceptance

criteria not already covered by an existing story?

1. Backlog / queue — Did priorities or P-tier status change in a way

that belongs in the tables (rare for a pure discussion thread)?

If none of the above apply, go to Exit.

Output format

MUST use these sections in this order:

1. Summary — One short paragraph: what (if anything) is worth capturing.

1. Proposals table (omit rows with no proposal):

| Artifact | Proposed path | Why (one line) | Follow | |----------|---------------|----------------|--------| | ADR / Research / Reference / Story / Backlog / Queue | e.g. {{DOCS_DIR}}/adr/004-foo.md | ... | link to relevant skill or docs overview |

1. Do not write files unless the user explicitly asks to create or update

them in the same turn. Offer drafts only when useful.

1. If proposing an ADR, remind: next NNN-, no renumbering; use

doccraft-adr.

1. If proposing a story, remind: use doccraft-story; consider updating

{{DOCS_DIR}}/backlog.md (status column) and {{DOCS_DIR}}/queue.md (pick-next order) when priorities shift, and invoke doccraft-queue-audit in the same turn if new depends_on edges were introduced.

Example output

**Summary:** This thread established a caching strategy for API responses
and identified a follow-up task for cache invalidation testing.

| Artifact | Proposed path | Why (one line) | Follow |
|----------|---------------|----------------|--------|
| ADR | `docs/adr/012-api-response-caching.md` | Decided on Redis with 5-min TTL over in-memory; worth freezing | `doccraft-adr` |
| Story | `docs/stories/p2-cache-invalidation-tests.md` | Bounded deliverable with clear acceptance criteria from discussion | `doccraft-story` |

- ADR: next number is 012. Use `doccraft-adr` to write.
- Story: use `doccraft-story`. If `depends_on` edges are added, invoke
  `doccraft-queue-audit` in the same turn.

Exit (nothing to capture)

If no row in the proposals table applies, respond with one or two sentences only, e.g. "No durable decisions or synthesis in this thread; nothing to add to {{DOCS_DIR}}/." — then stop (no filler table, no speculative ADRs).

Pre-execution validation

Before proposing any artifacts, MUST complete these checks:

1. Thread scan — review the entire visible conversation thread. NEVER

propose artifacts from assumptions about what was discussed.

1. Existing artifacts — check {{DOCS_DIR}}/adr/, {{DOCS_DIR}}/stories/,

{{DOCS_DIR}}/research/, and {{DOCS_DIR}}/reference/ for existing files that already cover the topic. Prefer updating existing docs over creating duplicates.

1. Config check — read doccraft.yaml to determine which categories

are enabled. NEVER propose a Research artifact when sessionWrap.capture.research is false, or a Business artifact when sessionWrap.capture.business is false.

1. ADR numbering — if proposing an ADR, scan {{DOCS_DIR}}/adr/NNN-*.md

to determine the next unused number. MUST include the correct next number in the proposal.

1. Story uniqueness — if proposing a story, scan

{{DOCS_DIR}}/stories/*.md frontmatter to confirm no existing story already covers the same scope.

Invalid examples (do not use)

• Proposing an ADR when the thread only explored options without reaching

a conclusion — use Research instead, or propose nothing.

• Proposing artifacts for a casual Q&A or debugging session with no

durable insight — use Exit instead.

• Writing files without the user explicitly asking in the same turn —

NEVER; propose only.

• Proposing a Business artifact when {{DOCS_DIR}}/business/ does not exist

and the project has not opted in via config — NEVER.

• Proposing speculative stories for work that was not discussed in the

thread — NEVER invent work to document.

• Including a filler table with empty rows when nothing warrants capture —

use the one-sentence Exit instead.

• Proposing a Reference doc when the content is a single sharp decision —

use an ADR instead.

• Proposing an ADR when the content is a broad comparison without a

conclusion — use Research instead.

Done condition

The task is complete when:

• The full conversation thread has been reviewed against the gate

checklist.

• Either a proposals table has been presented with concrete paths, one-line

justifications, and follow links for each proposed artifact, or the Exit response has been given.

• No files have been written (unless the user explicitly asked to create

or update them in the same turn).

• If an ADR was proposed, the next available NNN- number is included.
• If a story was proposed, a reminder to use doccraft-story and

consider doccraft-queue-audit is included.

Workflow reminders

• After the user approves a proposed ADR, invoke doccraft-adr to write

it — do not write the ADR manually.

• After the user approves a proposed story, invoke doccraft-story to

write it — do not write the story manually.

• If the user asks to create multiple proposed artifacts in one turn,

invoke each relevant skill in sequence.

• If the proposals table includes both an ADR and a story that references

it, create the ADR first so the story's adr_refs can reference it.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: alexpialetski
• Source: alexpialetski/doccraft
• License: MIT
• Homepage: https://www.npmjs.com/package/doccraft

Install and usage instructions live in the source repository linked above.