Add /lc-report skill: author MyST reports referencing ASTRA elements by path

[EiffL]

Context

PR #156 removed ASTRA's narrative field (astra-spec RFC-0002): rich write-ups now live as external reports that reference analysis elements by tree-path. PR #157 added the report scaffold to lc init (index.md + myst.yml wired to the MySTRA plugin) and explicitly deferred the authoring skills — this PR is that follow-up. Where we previously enforced building the narrative: block, the agent now naturally builds the MyST report.

New skill: /lc-report

claude/lightcone/skills/lc-report/ — authored against the MySTRA docs and Anthropic's skill best practices (specific third-person trigger description, SKILL.md as a concise overview, progressive disclosure into references/ one level deep, a validate→fix→rebuild feedback loop, single recommended path).

• SKILL.md (~100 lines): the three-layer model (data = spec/results, composition = report pages, presentation = theme); the golden rule (never hard-type a measured number, restate a decision, or re-describe an output — reference it by path); setup incl. creating the scaffold when lc init never ran (reproduction workdirs); mode selection; structure & multi-page along sub-analyses; draft order (methods → results → discussion → introduction last); coverage rule; myst build --html validation loop; restrictions & anti-patterns. Writes are locked to report pages + myst.yml.
• references/mystra-syntax.md: the exact MySTRA grammar — path grammar, {astra} role/directive (with options tables), {astra:ref}, {astra:cite}[:t], {astra:value} (col/where/pm/err/sig), anchors, registries, multi-page astra_scope, minimal myst.yml (with version-pinning note), build commands, and how every breakage renders ([mystra] warnings, error admonitions, ⟨value: …⟩ tokens).
• references/craft.md: the prose discipline ported from the retired narrative skill, re-expressed for the MyST surface — references-as-citations, decision vs. selection ({astra} vs {astra:value} on decisions), navigable data flow, length-as-compressor, real-subjects-real-verbs, anti-patterns.
• references/paper-reproduction.md: reproduction mode — paper→report mapping table, paraphrase/confidence-register/voice-seam rules, divergence reporting sourced from REVIEW's open-question resolutions, four-audit critique pass.

The three-mode model (paper reproduction / co-drafting / retrofit) survives from the old skill; the two former stub references are folded into SKILL.md as concise guidance instead of shipping near-empty files.

Wiring

• lc-from-paper: REVIEW close-out gains Step 4 — author the MyST report with /lc-report (steps renumbered; outputs, survey signals, phase table, and close-out summary updated). REPRODUCTION-SUMMARY.md stays the concise record; the MyST report is the rich, publication-shaped companion.
• lc-new (Done) and lc-from-code (Phase 3) now point the user at /lc-report at the natural moment.
• astra reference skill: the Description section notes that rich write-ups live in external MyST reports.
• session-start primer: drops the stale "narrative anchors" leftover from Remove ASTRA narrative field from skills; bump astra-tools to 0.2.10 #156 and primes /lc-report.
• Project CLAUDE.md templates: the inline _PROJECT_CLAUDE_MD and the lc init index.md stub mention /lc-report; the bundled claude/lightcone/templates/CLAUDE.md gains the ## Report section it was missing (it had diverged from the inline template).
• Docs: docs/skills/lc-report.md mirror, zensical.toml nav entry, rows in docs/skills/index.md + bundle README, and a /lc-report section in the user guide's agent-workflow page (which previously didn't mention the report at all).

Verification

• uv run pytest → 336 passed; ruff + mypy clean (the commands.py change is string-only and existing template tests still pass).
• just docs builds clean and emits /skills/lc-report/.
• SKILL.md frontmatter parses (name/description/allowed-tools).
• Repo-wide grep: no remaining ASTRA-field uses of "narrative" (leftover hits are general-English).

🤖 Generated with Claude Code

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Preview URL	Updated (UTC)
✅ Deployment successful! View logs	lightcone-cli	d3421b3	Commit Preview URL Branch Preview URL	Jul 05 2026, 04:57 PM

[github-actions]

✅ Eval Results

Metric	Value
Score	1.00
Build complete	✅
Cost	$2.51
Turns	83
Duration	508s
lightcone-cli	0.3.8.dev3+gbe2834805 (be283480)
Results	Download

Graders

✅ spec_valid (1.00)
✅ all_materialized (1.00)

Full output

16:57:22 lightcone.eval.build Building lightcone-cli wheel from /home/runner/work/lightcone-cli/lightcone-cli ...
16:57:27 lightcone.eval.build Built lightcone_cli-0.3.8.dev3+gbe2834805-py3-none-any.whl (commit be283480)
16:57:32 lightcone.eval.sandbox Created sandbox 21fd1208-6d51-4e59-b628-6d8fc30ed535 for trial build-snae-0
16:57:33 httpx HTTP Request: POST https://proxy.app.daytona.io/toolbox/21fd1208-6d51-4e59-b628-6d8fc30ed535/files/bulk-upload "HTTP/1.1 200 OK"
16:57:34 lightcone.eval.sandbox Installed wheels: ['lightcone_cli-0.3.8.dev3+gbe2834805-py3-none-any.whl']
16:57:35 httpx HTTP Request: POST https://proxy.app.daytona.io/toolbox/21fd1208-6d51-4e59-b628-6d8fc30ed535/files/bulk-upload "HTTP/1.1 200 OK"
16:57:35 httpx HTTP Request: POST https://proxy.app.daytona.io/toolbox/21fd1208-6d51-4e59-b628-6d8fc30ed535/files/bulk-upload "HTTP/1.1 200 OK"
16:57:35 httpx HTTP Request: POST https://proxy.app.daytona.io/toolbox/21fd1208-6d51-4e59-b628-6d8fc30ed535/files/bulk-upload "HTTP/1.1 200 OK"
16:57:36 httpx HTTP Request: POST https://proxy.app.daytona.io/toolbox/21fd1208-6d51-4e59-b628-6d8fc30ed535/files/bulk-upload "HTTP/1.1 200 OK"
17:06:05 lightcone.eval.sandbox Deleted sandbox for trial build-snae-0
  snae trial 0: score=1.00 complete

lightcone-cli: 0.3.8.dev3+gbe2834805 (HEAD be283480)
ASTRA: 0.2.10

  Eval Results: Scores  
┏━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Task ┃     Score     ┃
┡━━━━━━╇━━━━━━━━━━━━━━━┩
│ snae │ 1.00 +/- 0.00 │
│      │ pass@k: 100%  │
└──────┴───────────────┘

   Eval Results: Cost &   
         Duration         
┏━━━━━━┳━━━━━━━━━━━━━━━━━┓
┃ Task ┃ Cost / Duration ┃
┡━━━━━━╇━━━━━━━━━━━━━━━━━┩
│ snae │      $2.51      │
│      │      508s       │
└──────┴─────────────────┘

Total: 1 trials, $2.51, 508s

Results saved to: eval-results/build-be283480/results.json

[EiffL]

Closing in favor of re-proposing this on the new skills repo: the skill content will land in LightconeResearch/agent-skills stacked on #5 (the three-plugin migration). The lightcone-cli-side wiring here (lc init template + docs mentions of /lc-report) can be re-proposed separately once the skill lands there.