Doctor report: anthropics/skills@8a1541c

Public GitHub skill URL

https://github.com/anthropics/skills

Analyses requested

Follow-through analysis (skillspec doctor)
Security analysis (skillspec boundary)

Notes

Submitted from the public SkillSpec Doctor page.

Follow-through report — `skillspec doctor`

SkillSpec Doctor shape gate

Target: github.com/anthropics/skills

Shape: plugin_workspace

Skill files: 20
Plugin roots: 1
Decision: Run full plugin workspace assessment for 20 skill package(s) across 1 plugin namespace(s). Preserve namespace and path identity.

Shape-specific handling

This is a plugin-shaped workspace. The report must keep plugin namespace, folder path, repeated names, and repeated content as separate evidence instead of flattening them.

SkillSpec Doctor report

Target: github.com/anthropics/skills

Shape: plugin_workspace - plugin-shaped workspace with 1 plugin namespace(s); preserve namespaces before import

Status: workspace

What This Measures

Doctor estimates how likely an agent is to miss, reorder, improvise, use the wrong surface, or finish without proof because of the current skill shape.

Assessment scope: plugin workspace plus full per-package raw skill profiles; plugin namespace and skill path remain part of identity

Risk interpretation: Plugin workspace risk is the maximum of plugin-shape risk and per-package raw skill risk. Preserve plugin namespace/path identity and inspect package drift rows before import, compile, or install.

This is a baseline of the current skill shape at doctor time. It is not a grade of domain knowledge, human usefulness, author effort, or legal/medical correctness.

Higher risk means a higher chance the agent skips, reorders, or improvises load-bearing instructions.

Current Skill Baseline

Verdict: workspace risk: critical
Follow-through readiness: very low
Agent follow-through risk: critical (100/100, workspace aggregate)
Package risk rollup: max critical (100/100) at skills/algorithmic-art/SKILL.md; critical=14, high=5, medium=1, low=0

Shape Contract

Kind: plugin_workspace
Kind handling: plugin-shaped workspace; preserve plugin parent, namespace, shared files, and skills/ folder shape
Analysis status: workspace
Skill files: 20
Packages: 20
Namespaces: 2
Plugin roots: 1

Plugin Root Details

anthropic-agent-skills at `` with 19 skill file(s)

Next command

skillspec workspace map github.com/anthropics/skills --out <build-dir>/skillspec.workspace.yml

Surface

Activation body: 3868 line(s), 0 byte(s), approximately 68598 token(s)
Deferred resources: 20 file(s), 0 byte(s)
Frontmatter: 0 line(s), 0 byte(s)
Activation-loaded surface: 0%
Unmapped package files: 0
Instruction signals: 0 modal obligation(s), 0 late obligation(s), 0 numbered step(s), 0 dependency mention(s)
Code in active skill: 0 block(s), 0 unlabeled

Shape Details

Plugin Roots

anthropic-agent-skills at `` with 19 skill file(s)

Negative Signals

plugin markers require namespace-preserving workspace processing

Workspace Identity

Source files: 430 total
Namespaced packages: 20 package(s) across 2 namespace(s)
Skill content: 20 SKILL.md file(s), 20 unique byte content item(s), 0 repeated occurrence(s) in 0 referentiable group(s)
Estimated source tokens: approximately 71373 total if copied byte-for-byte, 71373 unique if referenced, 0 repeated
Frontmatter names: 0 repeated name group(s), 0 repeated occurrence(s)

Recommendation: Keep namespace/path as package identity and make repeated byte-identical SKILL.md content referentiable through source_content_ref aliases instead of copying the same bytes as independent source material.

Packages

anthropic-agent-skills:skills/academy-guide

Path: skills/academy-guide/SKILL.md
Role: plugin_skill
Drift risk: high
Discovery risk: low

anthropic-agent-skills:skills/algorithmic-art

Path: skills/algorithmic-art/SKILL.md
Role: plugin_skill
Drift risk: critical
Discovery risk: low

anthropic-agent-skills:skills/brand-guidelines

Path: skills/brand-guidelines/SKILL.md
Role: plugin_skill
Drift risk: high
Discovery risk: low

anthropic-agent-skills:skills/canvas-design

Path: skills/canvas-design/SKILL.md
Role: plugin_skill
Drift risk: critical
Discovery risk: low

anthropic-agent-skills:skills/claude-api

Path: skills/claude-api/SKILL.md
Role: plugin_skill
Drift risk: critical
Discovery risk: low

anthropic-agent-skills:skills/discernment-nudge

Path: skills/discernment-nudge/SKILL.md
Role: plugin_skill
Drift risk: critical
Discovery risk: low

anthropic-agent-skills:skills/doc-coauthoring

Path: skills/doc-coauthoring/SKILL.md
Role: plugin_skill
Drift risk: critical
Discovery risk: low

anthropic-agent-skills:skills/docx

Path: skills/docx/SKILL.md
Role: plugin_skill
Drift risk: critical
Discovery risk: low

anthropic-agent-skills:skills/frontend-design

Path: skills/frontend-design/SKILL.md
Role: plugin_skill
Drift risk: high
Discovery risk: low

anthropic-agent-skills:skills/internal-comms

Path: skills/internal-comms/SKILL.md
Role: plugin_skill
Drift risk: high
Discovery risk: low

anthropic-agent-skills:skills/mcp-builder

Path: skills/mcp-builder/SKILL.md
Role: plugin_skill
Drift risk: critical
Discovery risk: low

anthropic-agent-skills:skills/pdf

Path: skills/pdf/SKILL.md
Role: plugin_skill
Drift risk: critical
Discovery risk: low

8 more package(s) omitted. Use --json for the full list.

Findings

1. MEDIUM: Plugin-shaped workspace requires namespace preservation

Location: github.com/anthropics/skills

Evidence: Detected 1 plugin root(s) and 20 SKILL.md package(s).

Basis: contract_trace_static_well_formedness

Fix: Run skillspec workspace map; do not flatten skill names across plugin namespaces.

Next Actions

Recommended next action

skillspec workspace map github.com/anthropics/skills --out <build-dir>/skillspec.workspace.yml

Map the workspace before import: skillspec workspace map github.com/anthropics/skills --out <build-dir>/skillspec.workspace.yml.
From the harness, ask /skillspec map this repo and import the packages safely when you want the agent to preserve package shape, plugin namespaces, and dependencies.
After fanout import, read the converge/alignment reports: decision replay should pass, required package steps should be proven, and missing proof should be explicit.
Optionally publish the doctor baseline, generated skill.spec.yml files, and alignment reports with the source repo so reviewers can see what changed.
Restart the harness after install, then use the SkillSpec-backed skills normally rather than invoking internal CLI steps by hand.
Do not flatten the repo into one skill; process each discovered SKILL.md as an atomic package and converge dependencies before install.

Research Basis

17 cited basis item(s) are attached to this report.

reliability_gap_instruction_density - Large, dense activation bodies increase dropped-step risk and later instructions suffer primacy bias. (SkillSpec Skills Reliability Gap)
reliability_gap_metadata_context_pressure - Skill frontmatter metadata and activated bodies consume context; broad surfaces degrade routing and adherence. (SkillSpec Skills Reliability Gap)
reliability_gap_implicit_environment_contract - Scripts and snippets carry dependency contracts whether or not the skill declares them. (SkillSpec Skills Reliability Gap)
reliability_gap_no_execution_guarantees - Prose instructions and embedded snippets are guidance unless moved into checkable or enforced surfaces. (SkillSpec Skills Reliability Gap)
reliability_gap_unfilled_requirement - Reliable skills need a portable, machine-checkable account of intended behavior, dependencies, and proof. (SkillSpec Skills Reliability Gap)
contract_trace_activation_adherence_enforcement - Activation, adherence, and enforcement are separate gates; static prose mostly influences adherence, not enforcement. (SkillSpec Contract Trace Methodology)
11 more basis item(s) omitted. Use --json for the full basis registry.

Use --html for a shareable review page or --json for machine evidence.

Full artifacts: https://github.com/modiqo/skillspec/actions/runs/37062436287/artifacts/11251272781

The artifact contains: doctor-report.txt, doctor-report.md, doctor-report.html, doctor-report.json, shape-gate.md, shape-gate.json, source-stage.json, target.txt.

Security analysis — what each skill can reach

0 critical · 9 high · 0 medium · 4 low · 7 clean across 20 skill(s).
9 skill(s) reach beyond their own directory and warrant review before install.

Findings to review

HIGH — skills/canvas-design

🔴 analysis incomplete: additional effects may exist that this report and any derived policy do not cover. — canvas-fonts/ArsenalSC-Regular.ttf

HIGH — skills/claude-api

🔴 analysis incomplete: additional effects may exist that this report and any derived policy do not cover. — shared/preserved-thinking-migration/drop_block_probe.py
🔴 reads a secret from the environment → ANTHROPIC_API_KEY: can read a token or key held in the environment. — curl/examples.md:16
🔴 reads a secret from the environment → ANTHROPIC_AUTH_TOKEN: can read a token or key held in the environment. — shared/preserved-thinking-migration/drop_block_probe.py:404
🟠 agent-directing instruction: the agent could be steered away from what you asked. — SKILL.md:71
🟠 agent-directing instruction: the agent could be steered away from what you asked. — SKILL.md:35

HIGH — skills/docx

🔴 analysis incomplete: additional effects may exist that this report and any derived policy do not cover. — scripts/accept_changes.py
🔴 hidden content (conceal.zero_width): an instruction or payload a reviewer reading the skill never sees could act on the agent. — scripts/office/schemas/ecma/fouth-edition/opc-contentTypes.xsd:1
🟠 reads outside its directory → ../out.docx: can read files beyond the skill folder. — SKILL.md:56
🟠 writes outside its directory → ../out.docx: can modify files beyond the skill folder. — SKILL.md:56
🟠 runs a command chosen at runtime: what it runs cannot be determined before it runs. — scripts/accept_changes.py:68

HIGH — skills/mcp-builder

🔴 analysis incomplete: additional effects may exist that this report and any derived policy do not cover. — reference/node_mcp_server.md

HIGH — skills/pptx

🔴 analysis incomplete: additional effects may exist that this report and any derived policy do not cover. — SKILL.md
🔴 hidden content (conceal.zero_width): an instruction or payload a reviewer reading the skill never sees could act on the agent. — scripts/office/schemas/ecma/fouth-edition/opc-contentTypes.xsd:1
🟠 agent-directing instruction: the agent could be steered away from what you asked. — SKILL.md:34
🟠 reads outside its directory → ../out.pptx: can read files beyond the skill folder. — SKILL.md:62
🟠 writes outside its directory → ../out.pptx: can modify files beyond the skill folder. — SKILL.md:62
🟠 runs a command chosen at runtime: what it runs cannot be determined before it runs. — SKILL.md:57

HIGH — skills/skill-creator

🔴 analysis incomplete: additional effects may exist that this report and any derived policy do not cover. — eval-viewer/generate_review.py
🟠 writes outside its directory → /dev/null: can modify files beyond the skill folder. — SKILL.md:238
🟠 runs a command chosen at runtime: what it runs cannot be determined before it runs. — eval-viewer/generate_review.py:291

HIGH — skills/web-artifacts-builder

🔴 analysis incomplete: additional effects may exist that this report and any derived policy do not cover. — scripts/shadcn-components.tar.gz
🟠 writes outside its directory → /dev/null;: can modify files beyond the skill folder. — scripts/init-artifact.sh:34
🟠 runs a command chosen at runtime: what it runs cannot be determined before it runs. — scripts/init-artifact.sh:65

HIGH — skills/webapp-testing

🔴 analysis incomplete: additional effects may exist that this report and any derived policy do not cover. — scripts/with_server.py
🟠 writes outside its directory → /mnt/user-data/outputs/console.log: can modify files beyond the skill folder. — examples/console_logging.py:31
🟠 runs a command chosen at runtime: what it runs cannot be determined before it runs. — scripts/with_server.py:69

HIGH — skills/xlsx

🔴 analysis incomplete: additional effects may exist that this report and any derived policy do not cover. — scripts/office/soffice.py
🔴 hidden content (conceal.zero_width): an instruction or payload a reviewer reading the skill never sees could act on the agent. — scripts/office/schemas/ecma/fouth-edition/opc-contentTypes.xsd:1
🟠 runs a command chosen at runtime: what it runs cannot be determined before it runs. — scripts/office/soffice.py:68

Cleared: 7 clean (reach nothing outside their directory); 4 low (touch only their own files).

What to do next

1. Check and decide. Review the findings above, or re-run the analysis yourself:

skillspec boundary assess https://github.com/anthropics/skills

2. Install safely through the gate. Assess, then install only on your approval — across harnesses (Claude, Codex, or any AGENTS.md/SKILL.md skills dir):

skillspec pull https://github.com/anthropics/skills

3. Guard your harness. Compile a deny-by-default policy from the surface and enforce it on every tool call, so an unreviewed effect is blocked:

skillspec boundary emit <skill> --format skillspec   # least-privilege policy
skillspec boundary guard install                     # managed PreToolUse hook (observe mode)
skillspec boundary guard mode enforce                # deny effects no policy covers

Static analysis only — nothing in the package was executed. See the boundary guide.

Full analysis artifacts: https://github.com/modiqo/skillspec/actions/runs/37062436287/artifacts/11251272781

The artifact contains: boundary-report.md, boundary-assess.txt, boundary-assess.json.

Doctor report: anthropics/skills@8a1541c #58

Description