punitarani

@punitarani source repo

14 skills
0 followers
8 hours ago last updated

▌
Pr Ready · punitarani
Use when getting an abadge PR merge-ready, checking whether a branch is mergeable, resolving conflicts against main, recovering after a rebase, verifying CI is green, or shepherding a PR through checks and review comments. Triggers on "is this merge-ready", "get this PR green", "rebase onto main", "did I lose commits", "force push", "address the review comments".

0
installs
▌
CLI Release · punitarani bundle
Prepare, validate, and publish abadge CLI releases and the PRs that carry them. Use when updating the CLI release pipeline, checking changesets or versioning, dry-running release artifacts or the installer, or committing, pushing, reviewing, and merge-prepping the CLI release PR.

0
installs
▌
Abadge E2e Sweep · punitarani bundle
Use when the user wants to run, resume, monitor, or stop a long-running end-to-end test sweep of the abadge codebase (web, API, CLI, MCP, daemon, crypto, DB, SDK), including phrases like "sweep abadge", "run the e2e audit", "continue the test campaign", "resume the sweep", "what's the sweep finding", "stop the sweep", or any request to methodically test every surface of abadge in a loop with subagents and durable issue tracking.

0
installs
▌
Live Test Matrix · punitarani bundle
Define and execute a comprehensive end-to-end test matrix for an abadge feature against a live local stack — not just code-level integration tests, but real CLI binary invocations, real Hono+tRPC API calls on the wrangler emulator, real Better Auth sessions, real agent bearer tokens, and real Postgres state verification. Categorize the matrix into happy paths, edge cases, adversarial scenarios, and security pentests (≥3 variations per category), track every row in a `TESTING.md` running log, and execute via a generated bash harness. Use this skill whenever the user wants thorough manual or end-to-end testing of a feature, asks to "pentest" or "adversarially test" something, says "actually test it not just code tests" or "manually run the CLI against this", asks for a test matrix with multiple categories, or wants to verify a feature works against a real running stack. Prefer this skill over ad-hoc one-off testing scripts whenever the user wants more than three or four assertions, even if they don't explicitly

0
installs
▌
Abadge Security Audit · punitarani bundle
Use when the user wants to run, resume, monitor, or stop a deep, multi-wave security/compliance audit of the abadge codebase — code review, pen testing, threat modelling, and the full cybersecurity review pipeline. Triggers on phrases like "security audit abadge", "pen test the codebase", "start the security review", "continue the security audit", "what did the audit find", "generate the security report", "stop the audit", "production readiness security checklist", or any request to methodically audit all trust boundaries of abadge (api, web, sdk, cli, mcp, daemon, crypto, auth, db) in a loop with subagents, durable finding files, and honest saturation gating. READ-ONLY by contract — no code edits. Distinct from abadge-e2e-sweep, which tests functional correctness; this skill reasons about adversarial behaviour.

0
installs
▌
Authoring Graded Tasks · punitarani
Use when writing an eval task instruction, oracle, or grader over a simulated world - covers the brief, declaring the rule kind, structural floors, deliverable shape, and bounding the work. Load before writing instruction.md or a solver.

0
installs
▌
Measuring Model Limits · punitarani
Use when building or fixing an RL environment, eval task, or agent benchmark - the entry point that routes to world-building, task-authoring, gating, rollout analysis, and difficulty iteration. Enforces the rule that only a model failure may ship.

0
installs
▌
Validating Task Premises · punitarani bundle
Use when you are about to write an eval task, oracle, grader, or register against a generated world - the measure-first protocol that checks whether the pattern a task depends on actually exists, is evenly spread over time, is reachable by the agent through a tool, and discriminates. Load this before authoring a task from an idea of what a world "should" contain, whenever a measurement returns a suspiciously clean zero, whenever a graded field might be constant, and whenever a task's difficulty rests on a rate nobody has broken down by time.

0
installs
▌
Building Simulated Worlds · punitarani
Use when generating a simulated workplace, institution, or multi-agent history that tasks will be graded against - covers determinism, the offstage boundary, coherence gates, artifact realism, and fidelity measurement. Load before writing any world generator.

0
installs
▌
Iterating Task Difficulty · punitarani
Use when an eval task scores at ceiling or out of its target band and you need to move it - covers which difficulty levers are measured to do nothing, the coverage-versus-rule distinction, and which levers are forbidden. Load before changing a task to change its score.

0
installs
▌
Analyzing Rollout Failures · punitarani
Use when reading eval rollouts, trial logs, or trajectories to work out why a model scored below ceiling - covers what a zero actually means, DNF handling, sample size, and certifying a miss as a model failure rather than a task defect. Load before recording any score.

0
installs
▌
Gating Environment Quality · punitarani
Use when adding tests or gates that protect an eval suite's correctness - oracle independence, reachability, coherence, degeneracy, rule-phrasing, grading guards. Also covers falsifying a gate and auditing your own measurement tooling. Load before trusting any check you wrote.

0
installs
▌
Running Recorded Simulations · punitarani bundle
Use when running, supervising, resuming, or babysitting a long generative simulation or recording that takes hours to days - covers supervisor design, the resume-not-restart rule, what may and may not change while a run is live, and accepting on the artifact instead of the progress log. Reach for this whenever a run is measured in days of simulated time or hours of wall time, whenever you are tempted to restart something that died, and whenever you find a defect while a recording is still in flight.

0
installs
▌
Fli · punitarani
Guidance for installing and using Fli correctly as a CLI and MCP server. Use when: setting up Fli with pipx, running `fli` flight searches, configuring Claude Desktop with `fli-mcp`, using the HTTP MCP server, or troubleshooting Fli command availability and common usage mistakes.

0
installs

14 published skills