Agent Personas 0.5.0

• Add the Reverse Engineering Specialist persona and bundle: 7 core skills plus 4 conditional tool adapters (11 total). It works at the strongest analysis level available, from format tools to decompilers, integration layers and isolated runtime, and states which level it used.
• Reverse-engineering methods cover competing hypotheses with evidence for and against, information-gain-driven next steps, source-to-sink and pre-authorisation tracing, indirect-dispatch and library-boundary resolution, cross-version comparison and durable case state. When installed tools cannot answer a question, it searches for and evaluates new tools, then proposes them for approval instead of stopping.
• Expand the Frontend UX Engineer from 8 to 12 skills: art direction across ten style families with no default style, tool-agnostic image and video asset production, live render verification with independent review, and design comparison prototyping.
• Publish the 0.3.0 and 0.4.0 collections in this release; the full catalogue tables now list every persona, bundle and skill.
• Owner-accepted revisions without a further benchmark: evidence cards apply only to the versions they record, and no new uplift is claimed.

v0.2.1 — Personas and specialised skill bundles

Personas and bundles, together in the main catalogue

• Coding personas now appear alongside the other roles, rather than in a separate add-on section.
• Persona and bundle publication status is released. This is a normal release, not a prerelease; behavioural evidence and platform coverage remain documented separately.
• The README now starts with the supplied agent-persona illustration and explains the connection between agents, personas, specialised bundles and focused skills.
• A worked implementation example, role-selection table and complete clone/verify/stage instructions provide a clearer first-use path.
• Package identifiers and member selections are preserved. Engineering procedures are unchanged; current labels, version metadata and evidence hashes are reconciled.

Verification: 720 checks passed; GitHub Markdown rendering and the documented staging example passed. The image is included in the repository. No new model benchmark claims or live-profile installation are implied.

Get started · Catalogue · Evidence

v0.2.0 — Experimental coding specialists

Agent Personas — experimental coding collection

Three new complementary personas and bundles:

• Software Implementation Specialist
• Codebase Refactoring Specialist
• Adversarial Code Reviewer

The full catalogue now contains 117 skills, 23 personas and 24 bundles. Original instructions and consumer scripts are MIT licensed; source attribution and third-party notices are included.

Verified for this release

• 717 combined private-library and consumer regression checks passed.
• All six final disposable refactoring comparison runs passed acceptance. The release instructions ran meaningful compatibility tests before restructuring in two targeted cases; the baseline edited first. A trivial rename reused existing tests.
• All 42 new persona/bundle filesystem staging selections passed across the seven documented harness targets.
• Existing loadable packages and consumer scripts are unchanged.
• Publication audit checked the reachable history for known credential/private-binding patterns and the portable package inventory.

Limits

Experimental, not independently certified. Small synthetic tests on one model do not establish general accuracy or efficiency gains. Preparation took longer. Staging is not native agent execution certification. No live profiles are installed by this release.

Start with the README, then read the coding-specific evidence and harness compatibility.