CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

he-technical-review

Review diffs, PRs, specs, plans, or review-feedback items and return severity-ranked engineering findings with exact locations. Use when technical risks or feedback correctness must be verified before implementation.

Quality

78%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Low

Low-risk findings worth noting

Fix and improve this skill with Tessl

tessl review fix ./Plugins/harness-engineering/fixtures/budget-archive/2026-04-21/deferred-store/skills/code_quality_review/he-technical-review/SKILL.md
SKILL.md
Quality
Evals
Security

Quality

Content

78%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A well-organized, token-efficient review skill with a clear sequenced procedure and an unusually thorough Validation section. Its weaknesses are deferred detail that the bundle does not actually deliver: severity definitions, the findings template, and the canonical contract are all pointed at files missing from this bundle, and the one file that does exist (references/context-preservation.md) is not clearly linked.

Suggestions

Ship the referenced detail in the bundle or inline it: either include Infrastructure/references/contract.yaml and finding.md.tmpl, or inline a minimal findings-output example and one-line P0-P3 severity definitions directly in SKILL.md so reviewers can classify findings without external files.

Fix the References section to match the actual bundle: link references/context-preservation.md explicitly (or remove the 'compatibility mirror' pointer), and drop or correct paths to files that don't exist (contract.yaml, evals.yaml, task-profile.json, finding.md.tmpl, assets/icons).

Attach the three dangling 'Read when:' lines to the specific references they describe (e.g. domain-model-routing.md and qa-intake-routing.md) as bolded lead-ins, so each read-condition is unambiguously tied to its target file.

DimensionReasoningScore

Conciseness

The body is lean fragment-style throughout ("Findings first, implementation second.", "Resolve mode and target; stop if unusable.") with zero concept explanations Claude already knows and no padded sections. Every section (Philosophy through Subagent Routing) carries operational content, matching 'every token earns its place'; there is no over-explanation to trim, so anchor 4 does not apply.

5 / 5

Actionability

Concrete, executable instruction-style guidance: a 6-step procedure, an explicit output contract ("findings (`P0-P3`) with location, impact, minimal fix, confidence", "feedback_response_plan (`accept`, `clarify`, `push_back_with_evidence`)"), and concrete QA verification steps ("compare reported behavior, expected behavior, reproduction path, and proposed fix evidence"). Not a 5 because key details are deferred to files missing from the bundle — the finding template (finding.md.tmpl) and the canonical contract (Infrastructure/references/contract.yaml) — leaving no worked example of an actual finding; it clearly exceeds anchor 3's 'pseudocode / missing key details' level.

4 / 5

Workflow Clarity

The Procedure is a clear numbered sequence and the Validation section supplies explicit checkpoints ("Ensure mode matches target and findings contain severity, location, impact, minimal fix, confidence", "Fail fast: stop at first blocking prerequisite or failed validation gate"). Not a 5 because the severity scale itself is undefined in the body — P0-P3 are named but never characterized, so a reviewer cannot consistently classify findings without the missing contract file; not a 3 because validation gates are explicit, not merely implicit.

4 / 5

Progressive Disclosure

The body is well-sectioned, but scored against the actual bundle: the References section points to files that do not exist here (./Infrastructure/references/contract.yaml, ./finding.md.tmpl, ./assets/icon-*.png), while the one real bundle file, references/context-preservation.md, is never individually linked — it is only gestured at as a 'Compatibility mirror (non-canonical): ./references' directory pointer. The three trailing 'Read when:' lines are also ambiguously detached from their intended targets. That is 'references present but not clearly signaled / could be better organized' (anchor 3); it is not anchor 2 since the SKILL.md is not a monolithic inline dump.

3 / 5

Total

16

/

20

Passed

Description

71%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A solid description: concrete targets, a precise output contract, and an explicit 'Use when' clause in third-person voice. Its main weaknesses are a 'when' clause framed as abstract conditions rather than user-utterance triggers, and overlap risk with sibling code-review skills on the broad 'review diffs/PRs' surface.

Suggestions

Rewrite the 'when' clause around concrete user phrasings, e.g. 'Use when the user asks to review a PR/branch diff for correctness, validate review feedback or a CodeRabbit comment before implementing it, or check whether a proposed fix actually resolves a reported QA issue.'

Add natural synonyms to widen trigger coverage: 'branch', 'code review', 'technical soundness of a spec or plan', 'validate review feedback'.

Sharpen distinctiveness by leading with the niche (findings-first review with severity-ranked output and feedback verification) and moving the generic 'Review diffs, PRs' phrasing after it, so it is less likely to intercept plain code-review requests meant for he-code-review.

DimensionReasoningScore

Specificity

Concrete actions with a precise output contract: "Review diffs, PRs, specs, plans, or review-feedback items and return severity-ranked engineering findings with exact locations" enumerates several specific action-object pairs and pins down the deliverable. Not a 5 because only two verbs are used and capabilities covered in the body (e.g. domain-language drift checks) are absent; not a 3 because the enumeration and output specificity go beyond '1-2 concrete actions, not comprehensive'.

4 / 5

Completeness

Both parts are explicitly answered: the 'what' is "Review diffs, PRs, specs, plans, or review-feedback items and return severity-ranked engineering findings with exact locations", and the 'when' is the explicit clause "Use when technical risks or feedback correctness must be verified before implementation". Not a 5 because the 'when' is expressed as abstract conditions rather than concrete user trigger phrases (compare 'when the user mentions PDFs, forms'); it clearly exceeds anchor 3, where 'when' is missing or only weakly implied.

4 / 5

Trigger Term Quality

Natural terms users would say are present: "diffs", "PRs", "specs", "plans", "review", "feedback", "verified". Not a 5 because common variations are missing — no "branch", "code review", "technical soundness", or file/extension equivalents; not a 3 because the included keywords map well onto real user requests like 'review this PR diff' or 'verify this feedback'.

4 / 5

Distinctiveness Conflict Risk

The opening "Review diffs, PRs, specs, plans" is a broad surface shared with generic code-review skills, and the skill itself routes to a closely related sibling (he-code-review), so a user asking for a plain PR review could trigger this skill instead. The tail does carve a niche ("severity-ranked... findings", "feedback correctness... verified before implementation"), but that is 'somewhat specific with residual overlap' rather than the 'minor overlap risk' of anchor 4.

3 / 5

Total

15

/

20

Passed

Validation

87%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 14 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

metadata_version

'metadata.version' is missing

Warning

relative_links

Relative link issues: 6 missing, 3 deeper-than-1-level, 5 suspicious

Warning

Total

14

/

16

Passed

Repository

jscraik/Agent-Skills

Path

Plugins/harness-engineering/fixtures/budget-archive/2026-04-21/deferred-store/skills/code_quality_review/he-technical-review/SKILL.md

Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.