feat(launch): CodeGraph usage block in the entry file; --next prints the whole task

[protonspy]

Three changes, one branch.

--next was dropping the line below the checkbox

Reported against a real task:

- [x] 7.7 (Unit) Record the transform in the written file's provenance, so `trim` and `offset`
      stop being implicit steps inside `align` and `pack`

The parser was fine — Task.End and Detail already covered line 2. The loss was in display: taskLine prints Summary(width), the checkbox line only. That is right for a list and wrong for --next, which returns one task, so a description stopping mid-sentence sent the reader to the file — the exact cost the reading surface exists to remove.

--next now ignores --width and prints the checkbox line whole plus the continuation as it sits in the file. Listings still clip. Task.Continuation() is the raw lines with flags removed; Task.Detail stays the one-line collapse for table rows.

scc launch writes a CodeGraph usage block into CLAUDE.md/AGENTS.md

An index nobody knows how to query is an index nobody queries. Launch already indexes; it now also splices the block, the way it does RTK's. Three constraints, each tested:

• only once the binary is present — a block promising scc graph explore where that command fails costs the whole entry file its credibility, and CodeGraph is usually absent when init runs;
• before the index, so a failed index still leaves the agent knowing how to rebuild one;
• never on --json/--dry-run — a flag nobody expects to change anything must not edit a file the user owns.

The markers are scc's own (<!-- scc:codegraph-instructions v1 -->), deliberately the opposite of the RTK decision: sharing RTK's markers is what makes rtk init and scc rtk converge on one copy of a block they both write, while CodeGraph writes nothing into the entry file at all.

internal/mdblock

The marker-delimited splice, extracted so RTK and CodeGraph share it. Choosing the markers is the integration's decision; find/compare/replace/append and CRLF preservation are one implementation rather than two sets of edge cases. Its test pins the property that is actually new: two blocks coexist in one document and each updates without touching the other.

assets.Version is deliberately not bumped — codegraph.md is a fragment like rtk.md, not in Workspace(h) and not manifest-tracked, so a bump would change nothing for scc update.

CLAUDE.md gains a runnable "how to actually use it" section for scc graph, which was documented only as decisions and never as commands.

🤖 Generated with Claude Code

https://claude.ai/code/session_01U9ofAKiYuCCyUnqjfgyNkw