Conversation
- Suppress intermediate reasoning: add no-reasoning Critical note to copilot-cli delegation prompt - Avoid duplicate fetches: control center fetches shared data once, passes as context - Enforce plain text report format: ban bold/headers in delegation template - Fix⚠️ semantics in claude-cli: static analysis complete = ✅, not⚠️ - Add Synthesis Step template for consolidated control center output Issues 3 (model routing) and 7 (background delegation) parked to ~/Projects/xflow-plans/ Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
- Add Simple/Standard/Complex scale to orchestration skill - Simple tasks stay inline — cross-CLI overhead not justified - Updated decision tree checks complexity before task type - Native subagent tier made explicit in routing - Q&A phase for Complex tasks: 1-turn clarification before execution - Bump version to 0.5.4 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
📝 WalkthroughWalkthroughBumps plugin manifest version to 0.5.4 and updates documentation: adds task complexity classification for orchestration routing, a Q&A phase for Complex tasks, status semantics for analysis tasks, an enforced structured-report constraint for Copilot CLI delegation, duplicate-fetch avoidance, and a final synthesis step. Changes
Sequence Diagram(s)sequenceDiagram
participant User as User
participant CC as Control Center
participant Agent as Delegated Agent
participant Data as Shared Data Source
User->>CC: Submit task
CC->>CC: Classify task (Simple/Standard/Complex)
alt Simple
CC->>CC: Handle inline or use native subagent
CC-->>User: Synthesize & respond
else Standard
CC->>Data: Fetch shared data (once)
CC->>Agent: Delegate with full spec + [Context]
Agent-->>CC: Structured report (plain-text labels)
CC->>CC: Synthesize results & verdict
CC-->>User: Respond
else Complex
CC->>Data: Fetch shared data (once)
CC->>Agent: Delegate with full spec + [Before executing] Q&A block
Agent-->>CC: Answer Q&A
CC->>Agent: Refine delegation (if needed)
Agent-->>CC: Structured report
CC->>CC: Synthesize results & verdict (resolve ❌ subtasks first)
CC-->>User: Respond
end
Estimated code review effort🎯 2 (Simple) | ⏱️ ~10 minutes Possibly related PRs
Suggested labels
Poem
🚥 Pre-merge checks | ✅ 3✅ Passed checks (3 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment Tip You can make CodeRabbit's review stricter and more nitpicky using the `assertive` profile, if that's what you prefer.Change the |
namos2502
changed the title
Complexity-gates first (Simple stays inline), then routes by reason: platform-specific, context isolation, different model. Complexity level (Standard/Complex) now governs spec detail, not the routing decision itself. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
🧹 Nitpick comments (1)
skills/orchestration/SKILL.md (1)
64-64: Consider adding language specifiers to fenced code blocks.Three fenced code blocks are missing language specifiers (lines 64, 101, 175), triggering markdownlint MD040 warnings. While these are templates rather than executable code, adding
textormarkdownidentifiers would improve linter compliance and rendering consistency.📝 Proposed fix for linter compliance
Line 64 (decision tree):
-``` +```text New task →Line 101 (Q&A phase template):
-``` +```text [Before executing]: This task is complex. In your first response, list any questionsLine 175 (synthesis step template):
-``` +```text **[Task / PR title]**Also applies to: 101-101, 175-175
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@skills/orchestration/SKILL.md` at line 64, Add language specifiers to the three fenced code blocks in SKILL.md that are currently untyped: the decision-tree block (the "New task →" template), the Q&A phase template block (the block starting with "[Before executing]: This task is complex..."), and the synthesis step template block (the block starting with "**[Task / PR title]**"); update each opening fence to include a language token such as text or markdown (e.g., ```text) so the linter (MD040) stops flagging them and rendering is consistent.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Nitpick comments:
In `@skills/orchestration/SKILL.md`:
- Line 64: Add language specifiers to the three fenced code blocks in SKILL.md
that are currently untyped: the decision-tree block (the "New task →" template),
the Q&A phase template block (the block starting with "[Before executing]: This
task is complex..."), and the synthesis step template block (the block starting
with "**[Task / PR title]**"); update each opening fence to include a language
token such as text or markdown (e.g., ```text) so the linter (MD040) stops
flagging them and rendering is consistent.
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 5cdcd561-f37c-40a9-848c-3d3f953b23e1
📒 Files selected for processing (5)
.claude-plugin/plugin.jsonCHANGELOG.mdskills/agents/claude-cli/SKILL.mdskills/agents/copilot-cli/SKILL.mdskills/orchestration/SKILL.md
The table template forced a per-agent breakdown even for single-agent flows. Prose guidance lets the control center lead with what matters (issues list) and only add agent attribution when it's useful. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Removes standalone Synthesis Step section — guidance now lives as step 6 in the protocol where it belongs in the workflow sequence. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
- Clarify 'rewrite in your own words' vs surface raw output - Define severity tiers inline with what they mean - Add per-agent attribution guidance (only when useful) - Add partial failure handling: surface blocker, hold verdict Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
/reload-plugins does not remove already-loaded plugins — user must restart the session for the uninstall to take full effect. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
There was a problem hiding this comment.
Actionable comments posted: 2
🧹 Nitpick comments (1)
skills/orchestration/SKILL.md (1)
54-62: Clarify precedence between “Standard/Complex” and native-subagent routing.Line 54 can be read as “always cross-CLI for Standard/Complex,” while Line 67 says to prefer host-native subagents first. Please make precedence explicit in prose to match the decision tree.
Suggested wording tweak
-4. Task is **Standard or Complex** — the work inside the agent justifies the delegation overhead +4. Task is **Standard or Complex** — *after* checking for a suitable host-native subagent, the work may justify cross-CLI delegationAlso applies to: 63-76
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@skills/orchestration/SKILL.md` around lines 54 - 62, The guidance is ambiguous between the rule "Task is **Standard or Complex** — the work inside the agent justifies the delegation overhead" and the rule "The host has a native subagent that can handle it"; update the prose in the SKILL.md section containing these rules so it states explicit precedence: always prefer the host-native subagent when available and capable, and only choose cross-CLI agent delegation for Standard/Complex work if no appropriate native subagent exists or if the delegation provides clear additional capabilities; revise the sentences around the "Task is **Standard or Complex**" and the "Do NOT delegate when: ... The host has a native subagent that can handle it" lines (and the adjacent 63-76 rules) to make that precedence sentence explicit and unambiguous.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@skills/orchestration/SKILL.md`:
- Around line 64-76: The three fenced code blocks in SKILL.md (the block
starting with "New task →", the block starting with "[Task]: <one sentence —
what needs to be done>", and the block starting with "[Before executing]: This
task is complex...") are missing language identifiers; update each opening
triple-backtick to include a language tag (e.g., ```text) so markdownlint MD040
is satisfied, leaving the block contents unchanged and keeping the closing ```
as-is.
- Line 174: Update the Agent SKILL docs to explicitly prohibit re-fetching when
source data is provided in [Context]: for each SKILL (e.g., copilot-cli SKILL
and claude-cli SKILL) add a clear sentence such as "If the required source data
is present in [Context], do not re-fetch it; use the [Context] payload as the
single source of truth," and include a short example delegation prompt showing
passing [Context] and avoiding additional fetch calls; ensure the guidance
mentions both not issuing HTTP/file fetches and not embedding raw re-fetched
output in the delegation prompt.
---
Nitpick comments:
In `@skills/orchestration/SKILL.md`:
- Around line 54-62: The guidance is ambiguous between the rule "Task is
**Standard or Complex** — the work inside the agent justifies the delegation
overhead" and the rule "The host has a native subagent that can handle it";
update the prose in the SKILL.md section containing these rules so it states
explicit precedence: always prefer the host-native subagent when available and
capable, and only choose cross-CLI agent delegation for Standard/Complex work if
no appropriate native subagent exists or if the delegation provides clear
additional capabilities; revise the sentences around the "Task is **Standard or
Complex**" and the "Do NOT delegate when: ... The host has a native subagent
that can handle it" lines (and the adjacent 63-76 rules) to make that precedence
sentence explicit and unambiguous.
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: b166f721-f206-47ac-a1e4-14fd8dd55ee5
📒 Files selected for processing (2)
commands/cleanup.mdskills/orchestration/SKILL.md
✅ Files skipped from review due to trivial changes (1)
- commands/cleanup.md
Comment on lines
64
to
76
| ``` | ||
| New task → | ||
| Platform-specific? YES → cross-CLI | ||
| Context isolation / different model? YES → cross-CLI | ||
| Host native subagent available? YES → use it (faster, no auth needed) | ||
| Default → handle in current context | ||
| Simple? → handle inline (no cross-CLI) | ||
| Host native subagent available? → use it (faster, no auth needed) | ||
| Platform-specific (GitHub, Anthropic API, etc.)? → cross-CLI | ||
| Context isolation needed (verbose output, long subtask)? → cross-CLI | ||
| Different model needed? → cross-CLI | ||
| Default → handle inline | ||
|
|
||
| When going cross-CLI: | ||
| Standard → full spec (problem + acceptance criteria) | ||
| Complex → full spec + Q&A turn before execution | ||
| ``` |
There was a problem hiding this comment.
Add language identifiers to fenced code blocks (markdownlint MD040).
The fences starting at Line 64, Line 89, and Line 106 are missing language tags.
Lint-compliant patch
-```
+```text
New task →
Simple? → handle inline (no cross-CLI)
Host native subagent available? → use it (faster, no auth needed)
@@
-```
+```
-```
+```text
[Task]: <one sentence — what needs to be done>
[Context]: <only what the agent strictly needs>
@@
-```
+```
-```
+```text
[Before executing]: This task is complex. In your first response, list any questions
or ambiguities — 1 turn only. Do not perform any actions until I confirm.
-```
+```Also applies to: 89-100, 106-109
🧰 Tools
🪛 markdownlint-cli2 (0.21.0)
[warning] 64-64: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@skills/orchestration/SKILL.md` around lines 64 - 76, The three fenced code
blocks in SKILL.md (the block starting with "New task →", the block starting
with "[Task]: <one sentence — what needs to be done>", and the block starting
with "[Before executing]: This task is complex...") are missing language
identifiers; update each opening triple-backtick to include a language tag
(e.g., ```text) so markdownlint MD040 is satisfied, leaving the block contents
unchanged and keeping the closing ``` as-is.
|
|
||
| Fan out to multiple Agents in parallel when tasks are independent. Pass a report excerpt (SUMMARY + STEPS) as context into the next delegation prompt — never raw output. Agents do not chain to each other; all coordination happens at the control center. | ||
|
|
||
| **Avoid duplicate fetches:** When multiple agents need the same source data (e.g. a PR diff, a file's contents), fetch it once in the control center and pass it as `[Context]` in each delegation prompt. Do not let each agent re-fetch the same data independently. |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Verify whether agent SKILL docs explicitly enforce context-first / no-refetch behavior.
rg -n -C2 -g 'skills/agents/**/SKILL.md' \
-e 'Context' \
-e 're-fetch|refetch|fetch.*once|avoid duplicate|do not fetch'Repository: namos2502/xFlow
Length of output: 1032
Agent SKILL docs need explicit no-refetch guidance when source data is in [Context].
Agent SKILL docs (e.g., skills/agents/copilot-cli/SKILL.md:113 and skills/agents/claude-cli/SKILL.md:144) instruct passing [Context] in the next delegation prompt and not passing raw output, but they do not explicitly state that agents should not refetch source data when it is already available in [Context]. This gap creates ambiguity—agents may still redundantly fetch the same data without clear directive to avoid it.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@skills/orchestration/SKILL.md` at line 174, Update the Agent SKILL docs to
explicitly prohibit re-fetching when source data is provided in [Context]: for
each SKILL (e.g., copilot-cli SKILL and claude-cli SKILL) add a clear sentence
such as "If the required source data is present in [Context], do not re-fetch
it; use the [Context] payload as the single source of truth," and include a
short example delegation prompt showing passing [Context] and avoiding
additional fetch calls; ensure the guidance mentions both not issuing HTTP/file
fetches and not embedding raw re-fetched output in the delegation prompt.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This pull request introduces a new "task complexity" system to the orchestration skill, refines delegation logic, and clarifies reporting and prompt instructions for both the orchestration and agent skills. The changes aim to improve task routing, ensure appropriate spec detail, and optimize agent interactions—especially for complex or high-risk tasks.
Task Complexity and Delegation Logic:
skills/orchestration/SKILL.md, with clear signals and delegation styles for each level. This classification determines the level of detail in prompts and whether a Q&A phase is triggered before execution.Prompt and Reporting Protocols:
copilot-cli, delegation prompts must now always include a critical instruction to return only the structured report, with no extraneous reasoning or output.Control Center and Synthesis Improvements:
Documentation and Release:
CHANGELOG.mdfor version 0.5.4, summarizing the new features and protocol changes..claude-plugin/plugin.json.Summary by CodeRabbit
New Features
Documentation
Chores