justinedevs/meta-architect/sage

Sage

Use this skill inside Codex to verify or challenge stack choices with real sources. $sage remains the evidence gate even as Meta-Architect absorbs more native curation and reference packs.

Read .ma/context/technology-capability-matrix.json before making a stack recommendation. Treat each detected dependency as an inventory entry, not as proof of capability. Populate the entry's capability claims and technology-specific configuration variables only after mapping them to an official document or pinned upstream source. The generated .ma/specs/evidence.md must expose that inventory and its verification state so a reviewer can see which technologies were considered, which variables are project-wide, and which claims remain blocked.

Output

Produce:

• candidate tools, libraries, or services
• decision, status, evidence, blockers, next_allowed_triggers
• why each option fits or fails the architecture
• source-backed evidence from official docs, upstream repos, or approved GitMCP sources
• evidence grade and exact upstream mapping
• recommendation with tradeoffs
• unresolved gaps or missing evidence
• exact next trigger, usually $flow

Rules

• Follow this order:
• use known upstream repos and official docs first when they already exist
• use discovery accelerators to find or narrow candidates
• map the selected candidate back to an exact upstream repo
• verify against upstream repos and official docs before approving it
• Use references/source-selection.md when you need the native source-selection contract, evidence ladder, or packaged discovery rules.
• Use approved discovery accelerators when you need faster OSS candidate discovery:
• Ossium (https://ossium.live/home) for trending, curated, YC-backed, and GSoC-linked OSS discovery
• Trendshift (https://trendshift.io/) for rising GitHub engagement and topic momentum
• Dev Hunt (https://devhunt.org/) for newly launched developer tools
• Libraries.io (https://libraries.io/) for package/dependency metadata, with caution because its public data is scraped and not validated/curated for accuracy
• Open Hub (https://openhub.net/) for project activity, contributor, and comparison signals
• Open-source Projects (https://www.opensourceprojects.dev/) for curated OSS discovery and detailed project writeups
• Discovery accelerators are candidate-finding tools, not approval surfaces. The user-facing Meta-Architect product stays native even when evidence sources are external.
• Do not invent package capabilities or maturity claims.
• Prefer primary sources over summaries when validating technical details.
• Do not treat discovery listings alone as VERIFIED evidence; promote candidates to upstream repos and official docs before approving them.
• If the evidence is weak or contradictory, say so clearly and keep the recommendation conditional.

Procedure

1. Turn the architecture decision into concrete claims that need verification.
2. Prefer official documentation, upstream repositories, specifications, and release notes.
3. Check version, license, maintenance, security, compatibility, and operational constraints.
4. Map each source to the exact claim it supports and flag stale or conflicting evidence.
5. Update the technology capability matrix with the verified capability,

variable, compatibility, security, and migration claims.

1. Recommend one option with tradeoffs only when the matrix entry is verified,

then hand the validated decision to $flow.

Default evidence dossier

Use `docs/reference/technology-evidence.md` to keep each choice tied to an exact upstream source, version or commit, license, maintenance signal, compatibility constraint, security record, and migration cost. The dossier organizes evidence; it never turns a discovery listing into approval.

Quality bar

• Discovery listings are leads, never approval evidence.
• A source-backed fact is kept separate from an inference or recommendation.
• Version-sensitive claims include the date or version checked.
• A technology with unknown capabilities or variables remains blocked; package

presence alone never unlocks a recommendation.

Example

When evaluating an MCP SDK, verify its supported protocol revision, skill-extension support, deployment model, and production lifecycle from primary sources before selecting it.