meta.reviews.review_delegation

The repository’s sage-review-checklist skill instructs the main agent to start one review subagent for each anchor requiring re-evaluation. Each subagent is given the anchor ID and an initial prompt directing it to run cargo meta review start <anchor> and evaluate the printed inputs. The main agent supplies any extra instructions or source context as recorded review inputs.

The subagent uses cargo meta review read for additional source context and cargo meta review record to save its completed report. Only one subagent owns an anchor’s active directory at a time; reviews for different anchors may run in parallel. The main agent reconciles requested edits, coordinates changes, and repeats the start/read/record cycle for affected anchors after edits, then runs cargo meta check again. TODO/N/A evidence is reviewed under the same workflow.

Read in specification context · design/meta/reviews.md

Implementation

Implemented crates/sage-meta/src/review.rs:15–20

sage_meta::review · lib:sage_meta

#![allow(unused)]
fn main() {
// {anchor}`meta.reviews.review_delegation`
pub const PROMPT: &str = "Review this anchor as its assigned subagent. Enumerate ALL requirements, qualifications, edge cases and scenarios. Inspect all supplied implementation and test fragments and trace each aspect to implementation behavior and actual test assertions or justified TODO/N/A evidence. Passing tests and associations alone are insufficient. Follow helpers and capture every additional source dependency with cargo meta review read ANCHOR PATH [selector] before completing the report. Explain prerequisites for justified TODO tests; do not claim deferred behavior is implemented. Distinguish source inspection from commands actually executed. Write a JSON report containing reviewer, report, and requested_edits. Leave requested_edits empty only when every aspect is accounted for and no edits are needed. Otherwise list concrete changes, including missing tests or justified TODO tests with prerequisite comments. Run cargo meta review record ANCHOR REPORT_FILE to finish. After edits or stale-input rejection, run cargo meta review start ANCHOR and review the fresh inputs. The gate fails until a current report with no requested edits is recorded. Dependency capture is agent-declared.";

#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
#[serde(deny_unknown_fields)]
pub struct ReviewChecklist {

}

See more source ↗

Tests and review

Indexed tests and the implementation-completeness and test-adequacy review; no test run is implied.

Implemented sage_meta::review_skill_assigns_one_owner_and_requires_start_read_record

crates/sage-meta/tests/review.rs · test:review

#![allow(unused)]
fn main() {
#[test]
#[sage::test(meta.reviews.review_delegation)]
fn review_skill_assigns_one_owner_and_requires_start_read_record() {
    let skill = include_str!("../../../.agents/skills/sage-review-checklist/SKILL.md");
    // The skill is the behavior under test: check obligations within their
    // responsible role, while allowing Markdown line wrapping to change.
    let coordinator = skill
        .split("## Coordinator: delegate each failing anchor")
        .nth(1)
        .unwrap()
        .split("## Reviewer:")
        .next()
        .unwrap()
        .split_whitespace()
        .collect::<Vec<_>>()
        .join(" ");
    for obligation in [
        "one review subagent per anchor, including TODO/N/A anchors",
        "anchor ID and the initial instruction",
        "cargo meta review start ANCHOR",
        "review the printed inputs",
        "Only one agent owns an anchor's validation directory at a time",
        "different anchors may be reviewed concurrently",
        "Avoid supplying uncaptured source context",
        "instructions must be saved with `start --instructions`",
    ] {
        assert!(coordinator.contains(obligation), "missing: {obligation}");
    }
    let reviewer = skill
        .split("## Reviewer: capture initial context")
        .nth(1)
        .unwrap()
        .split("## Coordinator: follow up and check")
        .next()
        .unwrap();
    assert!(reviewer.contains("cargo meta review read ANCHOR path/to/file.rs"));
    assert!(reviewer.contains("cargo meta review record ANCHOR /tmp/review-report.json"));
    let follow_up = skill
        .split("## Coordinator: follow up and check")
        .nth(1)
        .unwrap()
        .split_whitespace()
        .collect::<Vec<_>>()
        .join(" ");
    assert!(follow_up.contains("Reconcile findings and coordinate requested changes"));
    assert!(follow_up.contains("After changes, delegate fresh start/read/record reviews"));
    assert!(follow_up.contains("for affected anchors"));
    assert!(follow_up.contains("`cargo meta check` again"));
}
}

See more source ↗

Review: Accepted · final_quality.

Reviewed the full meta.reviews.review_delegation claim, associated review.rs PROMPT implementation and the complete review_skill_assigns_one_owner_and_requires_start_read_record test. Captured and read the actual .agents/skills/sage-review-checklist/SKILL.md dependency. No behavior tests were executed; this is source inspection.

The claim is explicitly about what the repository skill instructs, not automatic spawning or enforcement of agent behavior. The coordinator section requires one review subagent per failing anchor including TODO/N/A, supplies the anchor ID with the instruction to start and inspect printed inputs, assigns one owner per validation directory while allowing concurrency across anchors, forbids uncaptured source context and saves extra coordinator instructions with start --instructions. Reviewer sections require fresh start, inspection of all initial fragments, capture of every additional source dependency with review read and completion through review record. The follow-up section assigns reconciliation/coordination to the main agent, requires fresh affected-anchor start/read/record cycles after changes and rerunning cargo meta check, and prohibits substituting a coordinator draft for a subagent's completed record. The associated PROMPT reinforces assigned-anchor scope, captured dependencies, complete reports, TODO/N/A explanation and restart after edits/staleness.

The test includes the real skill text and extracts coordinator, reviewer and follow-up role sections separately. Its normalized-whitespace assertions require each delegation/ownership/concurrency/instruction-capture obligation in the coordinator section, explicit read and record command examples in reviewer sections, and reconciliation, affected-anchor refresh and check-again instructions in the follow-up section. These are meaningful assertions for this documentary contract: they test responsible-role placement and required workflow rather than merely searching the whole file for isolated tool names. Markdown line wrapping is intentionally not semantic. There are no hidden test helpers beyond standard string operations and the captured include file.

All claim aspects are present in the skill and checked by the test or directly traced to the relevant role instructions. This review does not claim automated enforcement or that every agent execution complies solely because text exists. No associated TODO/N/A evidence, concrete defect or necessary edit was found. No source changes are requested.

Browse meta anchors

116 anchors

Anchor Implementation Tests

meta (116) 102 Implemented24 TODO 103 Implemented36 TODO

meta.anchors (22) 22 Implemented 21 Implemented12 TODO

anchor_pages Implemented ImplementedTODO

Every specification anchor has a stable page identifying the rule and showing its definition, context, implementation entries, and tests/evidence. The page includes Rust excerpts and access to surrounding source, and exposes each entry’s placeholder status and explanation. Multiple sites remain individually inspectable rather than being reduced to a single coverage badge.

Symposiun AGentic Environment (SAGE)

meta.reviews.review_delegation

Implementation

Tests and review