DWS Shared Execution Contract

This file only serves as an entry point when DWS, cross-product flows, URL pre-check, or ambiguous intent is involved. For clear single-product requests, directly use the corresponding

dingtalk-*

skill; product root skills that already embed a minimal execution contract do not need to fully read this file first.

Minimal DWS Execution Contract

Only use
```
dws
```
; for structured reads add
```
--format json
```
and judge based on the actual response.
Directly invoke known commands; if parameters/constraints/security are unclear, consult the leaf narrow Schema. Only read the known leaf Help once if the Schema is unavailable; fix
```
unknown flag
```
once using the same leaf Help. For
```
unknown command
```
, do not check Help: first prioritize explicit suggestions in the error, then explicit compatible entries in already-loaded Skill/reference; if none, report drift and stop — full Catalog is forbidden. Low-frequency references do not default to Help; root/parent/product Help is forbidden. When discovered, must execute or explain the blocker.
Do not guess commands/flags/fields/IDs/accounts/business facts; IDs come from real responses. If target has zero hits/multiple candidates/unclear type, disambiguate first; only optional time/display scope uses contract defaults; stop if required information is missing.
Parse/read/write the same profile, IDs do not cross organizations. For multiple accounts, only use the unique
```
isOrgCurrent=true
```
; otherwise user specifies — selecting first item, most recent login, or most recently used account is forbidden.
Do not output/log tokens, refresh tokens, appSecret, or webhook tokens; do not ask for injected authentication.
Writes must match explicit intent; confirmation is ultimately determined by the Runtime gate/Schema, and
```
--yes
```
is only added after confirmation.
Verify results after writing; never claim success based on exit code. Exit must include a final response distinguishing completed, partial, blocked, pending confirmation, or failed; preserve existing data and
```
complete/hasMore/stopReason/failures
```
.
Display timestamps in the session timezone, preserve original values if necessary.
For authentication/permission/profile/confirmation/unknown errors, only read the corresponding
```
dingtalk-shared
```
reference; do not consecutively guess alternative commands.

For international, overseas, or

.io

regional logins,

dws auth login --intl

must be executed (add

--device

in headless environments);

--intl

is only for login, subsequent business commands route automatically based on the selected profile.

Product or cross-product rules incrementally load on top of the minimal contract. When the user has clear product-content intent, intent takes precedence over URL form; for multi-account selection and cross-organization rules, read

../dingtalk-misc/references/profile.md

. Local files, product boundaries, and cross-product delivery rules are only loaded in corresponding tasks, preventing the global manual from being included in every single-product request.

Progressive Loading

Only read the files required for the current task; do not load all shared references at once:

Current Situation	Required Reading
Clear single product	Corresponding `../dingtalk-*/SKILL.md` ; do not read routing references
Generic DWS, product selection needed	routing.md
Cross-product, multi-step, summary, or report	workflow-routing.md
Input contains user-provided alidocs `/i/nodes/` URL or unverified nodeId (even with clear product intent)	url-patterns.md; normalize dlink first, then enter the target product
Input contains other alidocs, shanji, or other DingTalk URLs with unknown type	url-patterns.md
Product boundaries still hard to judge	Relevant sections of intent-guide.md
Authentication, global flags, or output format issues	global-reference.md
Command returned error	error-codes.md; only read the error-related section
`confirmation_required` / write confirmation	confirmation.md
Command discovery, Schema / `--compact` / `--all`	schema-usage.md
Suspected capability not supported	capability-limits.md
Batch/multi-source collection	conventions.md
Fixed short workflows	lite-catalog.md corresponding section

Product commands, scripts, and field details reside in corresponding product skills, not duplicated in

dingtalk-shared

Routing Order When This Skill Is the Entry Point

First identify clear product-content intent; clear intent directly selects the corresponding product. For user-provided alidocs
```
/i/nodes/
```
URL or unverified nodeId, still read
```
url-patterns.md
```
: first perform node type probing, and when
```
extension=dlink
```
, save
```
result.fileId
```
from
```
drive info
```
as the entry ID and pass it to
```
dws doc info
```
, then resolve the target hop-by-hop via
```
linkSourceInfo.nodeId
```
, and hand the final target ID to candidate products. Stable IDs already returned with real types in the current call chain can be reused directly.
Request includes multiple time-ordered steps, cross-product data passing, or summary reports: even if URL is identified, also read
```
workflow-routing.md
```
and combine needed product skills according to the action guide; the current release package does not include standalone scenario skills.
Request is single-product but product unclear: read
```
routing.md
```
, then explicitly read target product
```
SKILL.md
```
.
Boundaries like
```
doc/drive/wiki
```
,
```
aitable/sheet
```
,
```
calendar/minutes
```
remain unclear: only read the corresponding sections of
```
intent-guide.md
```
.
If still cannot determine, ask the user for clarification; do not guess product or commands.

Cross-Skill Execution

Relative
```
Read
```
links in the body are runtime dependencies;
```
metadata.requires.skills
```
does not auto-load them.
After selecting the target product, the target skill's commands, parameters, and risk rules take precedence.
Multi-step flows pass real return values in order; parallel read-only collection follows the corresponding workflow/reference, writes are serial by default with step-by-step verification.
Operations already inlined in product skills are executed directly; only read deeper references when encountering parameters or boundaries not covered by that skill.

Error Shortest Path

```
unknown command
```
/
```
unknown flag
```
: run corresponding level
```
--help
```
, fix using public flags, retry at most once; if command selection is uncertain, read schema-usage.md.
```
reason=confirmation_required
```
: handle per confirmation.md; do not treat as ordinary validation error or silently add
```
--yes
```
.
Authentication or permission errors: read corresponding sections of
```
global-reference.md
```
and
```
error-codes.md
```
.
Other errors: prioritize
```
retryable
```
,
```
retry_after_seconds
```
,
```
next_retry_at
```
,
```
hint
```
, and
```
actions
```
in the JSON error. Only retry following server pacing when
```
retryable=true
```
is explicit; if retry semantics are absent, use
```
--verbose
```
for diagnostics and stop — do not keep trying alternative commands.
Explicitly unsupported capability: state the boundary, do not bypass through other interfaces.

dingtalk-shared

NPX Install

Tags

SKILL.md Content (Chinese)