# Apply the original methods in the current host

The original Skill's research, argument, writing, figure, citation, formatting,
review and delivery methods remain operative scientific guidance. Read the full
relevant resource, including useful scientific detail inside tables and examples.
The current entrypoint is a route into these methods, not a replacement summary.

## Obtain the methods from the same task

Read the ordinary public task snapshot. Its advisory `method_guidance` identifies
resources selected from the actual stage, saved workflow and scene. It does not
record that the host has read them or that the paper has passed their checks.

Use the same profile and task with the installed launcher:

```text
python scripts/paperspine5_web.py host methods --profile-root "<same-profile>" --task-id "<same-task>" --stage draft
python scripts/paperspine5_web.py host methods --local --task-id "<same-task>" --stage draft --resource references/motivation-thread-writing.md --format text --max-lines 100
```

Read the current task and select methods once, then reuse the selected paths
while the request and applicable methods are unchanged. `--local` reads an exact
registered resource from this installed Skill without MCP or a task read. Task ID
is context only; its version, configuration and authority are not observed.
Read one applicable resource at a time. Text output is a
bounded, directly readable frame with a source SHA-256, range, source text,
application notes, next line and END marker. Follow the next line with
`--start-line <next> --expect-sha256 <same-source-hash>` until source EOF; a changed
source is reported instead of joining pages from different versions. Keep the
actual received ranges with the short continuation context when interrupted.
The default JSON interface remains available for programmatic use.

The entire text frame defaults to 8000 characters; lower `--max-chars` to fit
the host's tool-output allowance. If END or the declared text is missing, that
page was not fully received: request a smaller page or read the indicated local
file. A single oversized line yields an explicit fallback instead of a clipped
success. Neither END, EOF nor a recorded range proves the host understood or
applied the method. Read linked scientific dependencies
when relevant. Reuse available knowledge and valid outputs, not read markers alone.
After context loss reopen original passages needed for the next action, even when
unchanged; do not repeat all reading solely because another task event occurred.

`--stage` selects the work being considered without changing the public task.
The current stages are intake, research, contribution, evidence, draft, figure,
review and delivery. Workflows such as revise and transfer are not new stages.
Explicit `--focus` selects a conditional operation; it grants no authorization.
The complete resource catalog is [current-method-routing.json](current-method-routing.json).
If the helper is unavailable, use this catalog to open the same original files
with normal file tools. Keep the task and report the specific access problem.

Select conditional methods from the actual request and saved configuration:
`author-voice` when voice restoration is enabled, `translation` for requested
language conversion, `platform-review` for a supplied detector report,
`legacy-recovery` only for a historical Runner task, and `skill-maintenance`,
`telemetry` or `publishing` only for those explicitly requested operations. Normal
research, writing, figure and review methods are selected directly by stage.
Choose the stage of the work you are doing, even if the saved task is already at
delivery; selecting reading must not reset the task or discard completed work.

Select methods for the actual next action, not a ceremony at every state change.
Keep enough continuation context to resume the current reading and outputs;
ordinary notes are sufficient. Related tasks can share unchanged method knowledge,
but never inherit another paper's configuration, choices, facts or review verdict.
After required reading, execute the useful work rather than producing a second
summary of every rule. Combine overlapping evidence/figure/review notes and checks.
Revisit the original text when it changes or a concrete uncertainty requires it.

## Preserve scientific meaning when translating old execution wrappers

Read each resource's `current_application` together with its original text. These
notes identify the current use of its requirements and any specific superseded
wrapper. No paragraph should disappear merely because it mentions an old tool.

- Old phase/J-number sequences identify scientific work, not a requirement to
  restart the paper. Carry their research, argument, comparison, writing and QA
  methods into the matching current action. Original round-specific method files
  remain useful even when a separate four-round optimization task is cancelled.
- Runner tickets, Master attestations and old per-stage receipt formats become
  ordinary host planning, actual tool execution, source/claim/figure mapping and
  review of real files. Preserve the evidence, distinctions and checks expressed
  by their fields. Do not manufacture the old administrative objects to unlock
  local work. Use the original contract only when explicitly maintaining that
  legacy engine; it does not become a second current task authority.
- Read configuration, materials, choices and feedback through the same public
  task and its derived files. Use public decisions for real Web input, ordinary
  file tools for paper content, and public publication/delivery calls for actual
  outputs. Example file names are examples unless an active tool actually requires
  that path. Never hand-edit a derived state file to imitate user input.
- Research depth, hypothesis tests, extra experiments or training follow the saved
  research scope. In materials_only, preserve and use supplied results, including
  rich figures without raw arrays. Public literature and figure learning remain
  normal capabilities. A local input directory alone is not an offline instruction.
- Citation identity and point-of-use support remain required. Actual abstracts can
  support their stated bounded points; full-text reading is needed where the claim
  requires it. Visual learning requires actually seeing the figure. Do not replace
  these distinctions with a universal full-text dossier or an OA-only bibliography.
- Independent review requires a real separate reviewer context (a different Agent
  from the writer-host, which stays in the current conversation) and actual current
  files. Original role/persona methods guide that review without forcing a fixed
  role chain or invented criticism. A reviewer can find no issues; do not create
  a fake objection, mandatory revision, identity, score or PASS record to satisfy
  an old template. The author may repair known defects while review is unavailable,
  keeping that review limitation explicit.
- Voice restoration, translation, journal transfer, competition/report conventions,
  detector-specific diagnosis, updates, telemetry and public release apply only
  to the relevant requested work. Scientific content must remain accurate during
  stylistic changes. A route to an operation is not authorization to upload private
  material, send messages, publish, submit, pay or install software.
- Preserve previous work and recheck dependencies actually affected by a revision.
  A new unrelated task version or another figure's edit does not invalidate all
  figures, citations or reviews. An exported file, state flag, hash, schema result
  or numerical score is never by itself a scientific-quality pass.

Use the current semantic manuscript exporter and the target reference document
for venue formatting. Do not append the legacy `submission_check.py --fix-fonts`
normalization to that pipeline: it overwrites existing fonts and geometry. The
current Word guard preserves styles and can inspect an explicit citation style;
numeric citations alone do not specify their punctuation.

For a justified nonwhite figure, inspect the actual design and contrast. The
visual-readiness prober supports a declared solid-color exception; the separate
legacy figure-reference checker still assumes a neutral canvas. Do not invoke
that old assumption as the current scientific acceptance rule or relabel a
nonwhite result as white to satisfy it. Other designs require actual visual review.

## Make the quality requirements visible in the work

At intake and research, preserve the full material/result inventory, saved scope,
data provenance, current literature and genre/venue requirements. At contribution
and evidence, connect the closest prior work, the actual research increment,
claim support, uncertainty and the complete set of useful supplied results.

At draft and figure work, apply the original argument-transfer, paragraph/section,
figure-story, reference-deconstruction and manuscript-assembly methods. Plan the
whole paper's visual argument before narrowing Results to an early summary chart.
Retain meaningful complexity, strong comparison structures and explanatory
mechanisms supported by the study or cited knowledge. Keep real references, actual
sources and editable outputs; distinguish figure alternatives from learning aids.

The figure story begins with result coverage and the scientific jobs of the
figures, not a predetermined chart count or convenient plotting primitive.
External-reference learning and old/new regression comparison both apply where
relevant. The current figure-story method preserves the panel/claim/source
relationships without requiring old Master attestations or Runner handoffs.

At review, the real independent reviewer obtains the applicable original review
methods from this same installed Skill. Review research substance, contribution,
reasoning, coverage, literature learning, prose, figure design and target format
alongside numeric/citation/render checks. Locate gaps in current passages, figures
or outputs; inspect what was actually produced against relevant exemplars. Report
validity, editorial quality, technical delivery and submission readiness separately.

At delivery and revision, check package contents, editable dependencies, authorized
sharing scope, actual browser-saved bytes, feedback and affected-work reuse. Keep a
short scientific issue-to-change record in existing notes. The method catalog and
paragraph audit are aids to finding and applying requirements; they do not replace
the scientific work or the reviewer's examination of its results.