fix(skills): preserve inline task text

[nightt5879]

Summary

• dispatch trailing task text in the same turn for $<skill> do X, /<skill> do X, and /skill <skill> do X
• keep bare $<skill> activation armed for the next user message
• preserve /skill install as the management command while allowing a literal install skill through $install ... and /install ...
• carry the active skill through immediate, queued, and steer message paths exactly once

Root cause

The unified skill command path accepted trailing arguments but discarded them before activation. The /skill <name> <task> path also treated the whole raw argument string as the skill name, so it could not both select the skill and dispatch the task.

Impact

Inline skill invocations now behave as a single request instead of silently dropping the user's task. Existing bare activation and skill-management behavior remain compatible.

Validation

• cargo fmt --all -- --check
• cargo test -p codewhale-tui --bin codewhale-tui --locked commands::groups::skills::skills::tests -- --test-threads=1 (27 passed)
• cargo test -p codewhale-tui --bin codewhale-tui --locked inline_skill_request_keeps_instruction_when_busy_queueing -- --test-threads=1 (1 passed)
• CI-equivalent cargo clippy --workspace --all-features --locked -- -D warnings ... passed
• cargo test --workspace --all-features --locked reached 6667 passed / 3 parallel-state failures; all three failures passed individually with --test-threads=1
• cargo build --release --locked -p codewhale-cli -p codewhale-tui; both binaries start and report version 0.8.68
• offline eval harness succeeded (6 steps, 0 tool errors)
• version drift, provider registry, changelog sync, diff, Cargo.lock, and co-author credit checks passed

Closes #3915

[nightt5879]

@codex review

Please review final head 66a6edf, including inline task dispatch, bare activation compatibility, the exact-name install skill collision, and queued skill delivery.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. 👍

Reviewed commit: 66a6edf5bb

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Hmbown]

Thank you, @nightt5879 and @CCChisato — merged for v0.8.68 with the original commit and co-author credit preserved. The release candidate keeps the additional queue-delivery regression and records both contributors in the release notes.