KhurrumMahmood

@khurrummahmood source repo

79 skills
0 followers
8 hours ago last updated

▌
Decide · khurrummahmood bundle
Author or amend an Architectural Decision Record under ai-docs/decisions/. Smallest first-class artifact in the senior-engineer skill ecosystem; used at any tier (Quick / Feature / System / Maintenance) when a real choice is being made. Reads canonical-patterns.md and architectural-smells.md to suggest related-pattern / related-smell backrefs and avoid duplicating an existing decision. Read-only against the rest of the codebase — only writes under ai-docs/decisions/. Hands off to /plan-feature, /scope-feature, or implementation work.

0
installs
▌
Orient · khurrummahmood bundle
Establish (or re-confirm) a project's lifecycle state on two independent axes — maturity (prototype → first-users → production) and stakes/exposure (internal → external → public-adversarial) — by asking orientation questions, showing the resulting classification and which standard rungs it activates, then writing the declared state to `.engineering/project-state.json` (the cross-agent state home, ADR 0021). This is the human-in-the-loop "pull" mechanism of ADR 0020 — automatic inference proposes a transition, /orient lets the human dispose. Stakes/exposure are NOT reliably inferable from code, which is why the questions matter. Idempotent — re-running re-confirms and overwrites.

0
installs
▌
Converge · khurrummahmood
After-phase convergence gate. Models the work-in-progress as a small value graph (every dimension must hold at once for "done"), then emits a fixed verdict — phase_status (advance/repair/branch/park/discard), strongest/weakest nodes, the single next necessary step, why that step, the success gate, the stop condition, and an explicit do-not-do-next list. Routes effort toward the next NECESSARY move instead of the next available one; refuses fan-out. Advisory and read-only against production code.

0
installs
▌
Diagnose · khurrummahmood bundle
Build a trustworthy debugging loop for hard bugs, regressions, and intermittent failures before changing code. Use when a symptom is unclear, a bug has not been reproduced, a performance regression needs measurement, or a fix would otherwise be based on guesswork.

0
installs
▌
Gut Check · khurrummahmood
Instinctive senior-engineer "does this look dumb?" reaction pass over a plan markdown, a commit diff, or a free-form architecture summary. Emits 3-5 cited smell reactions split into un-decided smells (no precedent / ADR covers them — raw signal) and decided-but-still-smell (an ADR or precedent explains why the project did it this way, but the smell still surfaces — surface both so the human can decide whether to re-litigate). Confidence-banded (strong-smell / weak-smell / style-preference). Prompt-only, no helper script.

0
installs
▌
Move Path · khurrummahmood bundle
Deterministically plan, dry-run, apply, and verify standalone path moves while updating identity-resolved Markdown, HTML, config, backtick, and exact path references. Checked JavaScript updates bounded literal module references; checked Go supports one leaf non-main package-directory move in one root module; checked Java supports one leaf package-directory move with compiler-attributed package/import/FQCN edits; checked Kotlin supports one same-package authored .kt source-location move with compiler-proved JVM identity; checked C# supports one same-namespace, same-filename authored .cs implementation-file location move in one strict dependency-free net10.0 executable project; checked PHP supports one Composer PSR-4 leaf namespace-directory move; checked Swift supports one dependency-free SwiftPM target-directory move while retaining module identity; checked Rust supports one conventional Cargo library leaf-module file or directory move; checked Ruby supports one evidence-authorized plain-gem module file move

0
installs
▌
Plan Spec · khurrummahmood
Final skill in the System-tier chain. Reads an `architected`-status plan, validates that §6 has no unresolved P0 forks, then calls `scripts/plans.py promote` to scaffold a successor spec under `ai-docs/specs/<spec-id>.md` and mark the plan `promoted`. The plan's §3-5 inform the spec's Architecture / Implementation / Exceptions sections. Hands off to `/refactor-subsystem` for execution.

0
installs
▌
Plan Skill · khurrummahmood
Plan and harden a new or revised skill before implementation, including adversarial requirements pushback, trigger design, evidence contracts, dogfood cases, and review gates. Use when creating a new .claude/skills skill or materially changing an existing skill.

0
installs
▌
Track Idea · khurrummahmood bundle
Append an intake, event, or lesson record to the Tier 1 idea ledger at .claude/ideas/log.jsonl. The capture surface for everything that could be reused or forgotten — features, refactors, prompts, harnesses, research probes, workflow tweaks. Validates the record against the schema in .claude/docs/idea-ledger.md before writing, then projects the idea's current state. Read .claude/docs/idea-ledger.md when authoring or debugging this skill.

0
installs
▌
Triage Debt · khurrummahmood bundle
Cross-cutting aggregator that scores accumulated debt across the maintenance loop's outputs (find-* report stacks, spec drift, decision drift, hard-size-overflow specs, recurring same-target hits) and produces a ranked queue at `reports/triage-debt/scan-<TS>/queue.md`. Each entry names the recommended next skill to invoke (refactor-subsystem / fix-workflow / extract-* / decide). Top-5 highlighted. Pure read — never edits production code, never runs find-*; reads the cached evidence those skills already produced.

0
installs
▌
Which Shape · khurrummahmood bundle
Recommend the right problem-solving loop before choosing individual skills. Reads the explicit shape registry plus project adapter/profile state, then returns an advisory route such as project-intake, bug-fix, concept-rename, or task-closeout (illustrative — shapes.json is the sole shape inventory). Use when the user describes a messy situation and should not need to understand the skill catalog.

0
installs
▌
Which Skill · khurrummahmood bundle
Recommend a skill (or "no planning skill applies — proceed directly") for a free-text task description. Reads a bundled metadata-only catalog without loading skill bodies, then emits on-demand guide/tool paths and a bounded fresh-sub-agent handoff. Ambient installation is explicit and optional.

0
installs
▌
Meta Audit · khurrummahmood bundle
Disinterested meta-failure audit of a plan, methodology, conclusion, or measurement system. Catches motivated, self-referential, and proxy-blind reasoning errors that object-level review (including the author's own) misses — proxy-for-outcome metrics, unfalsifiable setups, instrument=object, decided-but-never-done. Pass an artifact path or inline text. Six lenses + a novel-failure critic. Produces signal, not verdict.

0
installs
▌
Explain Code · khurrummahmood bundle
Read-only EXPLAIN skill that converts a Python, Go, Java, bounded Kotlin/JVM, JavaScript-family, TypeScript/TSX, PHP, Ruby, Swift, Rust, or Dart target's direct public declarations into an annotated behavior doc at reports/explanations/<target>.md. Unresolved surfaces remain visible instead of being inferred.

0
installs
▌
Extract Enum · khurrummahmood bundle
Turn a confirmed Django string-state field into a TextChoices proposal, a reviewed TypeScript or checked-JavaScript result into an as-const value-object proposal, or accepted Go, Java 17, Kotlin/JVM, PHP, Ruby, bounded Rust, Dart, or Swift implicit-state evidence into a typed-constant/enum proposal. Emits a caller and boundary inventory without editing production code.

0
installs
▌
Find Dormant · khurrummahmood bundle
Detect dead and quasi-dead code without changing source. Python retains vulture, AST, URL, silent-catch, and scout verification stages; TypeScript/TSX, checked JavaScript, Go, Java, bounded Kotlin/JVM, PHP, Ruby, Rust, Dart, and Swift have narrow static review branches for human review. Never infers safe deletion from static evidence.

0
installs
▌
Find Omnibus · khurrummahmood bundle
Detect omnibus modules — files answering questions from 3+ independently-understandable domains. Uses family-local syntax facts for Python, JavaScript/TypeScript, Go, Java, bounded Kotlin/JVM, PHP, Ruby, Rust, Dart, and Swift; then groups symbols by head-noun cluster, requires scout judgment, and produces decomposition evidence. Never edits code.

0
installs
▌
Fix Workflow · khurrummahmood bundle
Execute a cleanup surfaced by /find-duplication, /find-dormant, or /find-semantic-duplication. Accepts a cluster ID from a triage report (e.g. `cluster:P0-1`, `delete:foo`, `semantic:SC-1`), a file path, or a raw natural-language description. Loads context, writes a regression test first, refactors in a behavior-preserving commit, adds a separate bug-fix commit if latent bugs surface, runs the verification test matrix, writes a cluster learnings entry, and recommends the next cluster. Runs in the current worktree with full commit discipline.

0
installs
▌
Introduce Fk · khurrummahmood bundle
Turn a tuple-inferred-identity pattern (`.filter(status=X, created_at__gt=Y).first()`) into an explicit ForeignKey + two-step data migration proposal. Consumes an introduce-fk candidate from /find-implicit-state (or an explicit `OWNER_FILE::OwnerModel -> TARGET_FILE::TargetModel [via fk_name]` target) and emits reports/introduce-fk/<target>/proposal.md with the FK field shape, backfill migration sketch, caller migration table, tie-break strategy, and risks. Read-only — no code edits. Hands off to /fix-workflow or /refactor-subsystem.

0
installs
▌
Plan Feature · khurrummahmood bundle
Plan a Feature-tier change (1-3 day scope, touches one workflow, needs impact analysis but not new subsystem). Reads subsystem/workflow docs, canonical patterns, architectural smells, and the decision registry; fans out scouts per touched subsystem to map call sites, model touchpoints, and behaviors-to-preserve; surfaces decision stubs for material forks; and scaffolds a proposed-lifecycle spec under `ai-docs/specs/<feature-name>.md` so implementation work can resume from the spec.

0
installs
▌
Repair Skill · khurrummahmood bundle
Repair an existing skill whose text cannot be blind-trusted at execution time. Runs the validated five-role loop — frame review, claim-verifying scout, change spec with declared verdict, no-invention implementer, independent verifier — plus scaled A/B lift probes and (for execution-heavy skills) a real-host dogfood. Use when a frame review, dogfood, or production incident shows a skill producing wrong execution despite plausible text.

0
installs
▌
Adapt Project · khurrummahmood bundle
Discover objective host-project facts and scaffold a project adapter for engineering-skills. Reads Python, JavaScript-family, TypeScript, Go, Java, bounded Kotlin/JVM, PHP/Composer, bounded plain-Ruby/Bundler, SwiftPM, Cargo/Rust, and plain-Dart stack/source markers plus commands, tests, CI, docs, domain terms, sensitive surfaces, existing guardrails, and skill overlays; writes adapter artifacts under reports/adapt-project/scan-<TS>/ by default. Host writes to .engineering/project/adapter.yml require --apply, and --no-host-write is the dogfood mode for evaluating another project without touching it.

0
installs
▌
Engineer Init · khurrummahmood
Bootstrap the engineering-skills runtime so the script-backed skills can actually run. Health-checks candidate Python >= 3.11 interpreters, creates the .venv, installs requirements.txt plus pinned package-lock Node tooling when declared, wires pre-commit hooks when the repo is git-tracked, and verifies the install by running script-backed paths. Idempotent — safe to re-run; each stage skips work already done. Run once per clone, or whenever a skill fails with a missing-module / dependency error. Does not edit production code, does not git-init, and does not scaffold a new project (that is /init-project).

0
installs
▌
Map Subsystem · khurrummahmood bundle
Produce or refresh a durable inventory doc for a Python, TypeScript/TSX, checked-JavaScript, Go, bounded Java, Kotlin/JVM, C# 14/.NET 10, Composer PSR-4 PHP, dependency-free SwiftPM, compile-database-backed C/C++, plain locked Ruby gem, Cargo-backed Rust, or bounded plain-Dart subsystem at .engineering/docs/subsystems/<name>.md. Python covers file list, public surface, responsibility table, dependency graph, and convention-compliance score; language branches use family-local native attribution for their bounded facts. No refactor intent — MAP skill in the maintenance nervous system.

0
installs
▌
Scope Feature · khurrummahmood bundle
First skill in the System-tier chain. Scaffolds an `ai-docs/plans/<name>.md` plan if needed, reads decisions / canonical-patterns / smells, then drives clarifying questions to fill §1 (Scope & Bounds) and §2 (Success Criteria) of the plan. Advances plan status to `scoped`. Designed for System-tier work — new subsystems, cross-subsystem features, multi-week initiatives — where the judgment pause between scoping and impact analysis is the whole point.

0
installs
▌
Teach Pattern · khurrummahmood
Cross-cutting teach skill — converts a topic (smell name, decision id, subsystem name, canonical-pattern anchor, or free-form rule) into a layered briefing at `reports/teach-pattern/scan-<TS>/<topic>.md`. Five-section structure — rule (one line), why (smell + decision link), exemplar (real spec/decision/code reference), counter-example (real find-* finding), enforcement (lint or "none yet — file a /decide"). The `--for-agent` flag re-frames the briefing as "given THIS PR context, here's why X is right and Y is wrong" so an agent can use the output directly mid-task.

0
installs
▌
Unify Shadows · khurrummahmood bundle
Turn one confirmed semantic-duplication finding into an evidence-cited, read-only proposal. Python retains the profile-scout workflow; TypeScript/TSX, checked-JavaScript, Go, Java, Kotlin/JVM, PHP, Ruby, bounded Rust, Dart, and Swift consume accepted structured evidence and emit proposal.md, evidence.json, and scope.json with source/caller impact, native tests, stop conditions, human approval, and an honest template for all four consolidation shapes.

0
installs
▌
Which Cleanup · khurrummahmood bundle
Route a completed change to proportionate cleanup and guard skills. Use after editing or committing code to inspect only the changed scope, choose warranted closeout guides, and return on-demand guide/tool paths for direct or fresh-sub-agent execution. Ambient installation is explicit and optional.

0
installs
▌
Impact Feature · khurrummahmood bundle
Second skill in the System-tier chain. Reads a `scoped`-status plan, fans out one scout per touched subsystem to map call sites + behaviors-to-preserve, synthesizes the cross-subsystem impact + blast radius, and fills §3-4 of the plan. Advances plan status to `impacted`. Heavier than /scope-feature (scout fan-out) but lighter than /plan-feature (no spec scaffold; that's /plan-spec).

0
installs
▌
Query Patterns · khurrummahmood bundle
Search the Tier 2 pattern library at .claude/patterns/ for prior art that matches a problem description. Reads every pattern file's frontmatter (problem_class, pros, cons, domain, status, generalizability) plus the body's "Problem fit" / "Use this when X" headline, scores by token overlap, and returns ranked matches. Read-only. Read .claude/docs/pattern-library.md when authoring or debugging this skill.

0
installs
▌
Rename Concept · khurrummahmood bundle
Assess a domain-concept rename against its full lifecycle and a mandatory two-band completeness gate, read-only. Renaming a concept that lives in the glossary spans many surfaces — identifiers, docs, the cross-tool agent mirrors — and is executed from tribal knowledge with no front door and no completeness gate, so renames land half-applied. This skill (v0, assess-only) reports the scope-gate verdict, blast radius, a per-step lifecycle status table, and the completeness gate. It performs strict lexical assessment of Python, JavaScript, TypeScript, TSX, Go, Java, bounded Kotlin/JVM, PHP, Ruby, Rust, Dart, and Swift text. For `.ts`, `.tsx`, and checked `.js`/`.jsx`/`.mjs`/`.cjs`, a host-pinned TypeScript Compiler API resolves glossary identifier candidates, declarations, and references; it does not claim whole-project type-checking, JSX runtime behavior, or codemod safety. For Go, Go 1.22+ `go/types` classifies project-exported concept declarations and their references while retaining fields, import aliases, lo

0
installs
▌
Audit Decisions · khurrummahmood bundle
Read-only, portable decision-registry drift audit. It writes a final drift report, captures registry/link diagnostics, and validates `decision:NNNN` references from Python, Go, Java JDK 17+, bounded Kotlin/JVM, PHP, Ruby, Swift, Rust, and Dart, JavaScript-family, and TypeScript comments plus Markdown/HTML references.

0
installs
▌
Design It Twice · khurrummahmood
Resolve a material design fork by parallel-divergent fan-out — spawn 3 designer sub-agents with deliberately different binding constraints, then synthesize a comparative analysis with a recommendation. Use when a P0 fork from /architecture-fit (or any planning skill) has 2+ defensible answers and the cost of being wrong is high relative to the cost of fanning out. Writes one comparative-design document under reports/design-it-twice/scan-<TS>/; does NOT author the ADR — /decide is the downstream step.

0
installs
▌
Architecture Fit · khurrummahmood
Third skill in the System-tier chain. Reads an `impacted`-status plan, walks the impact map against the decision registry, canonical-patterns, and architectural-smells; surfaces every material fork that needs an ADR (suggests `/decide` candidates inline); fills §5 (Architecture Fit) and §6 (Open Decisions) of the plan. Advances plan status to `architected`. The last judgment pause before promotion to spec.

0
installs
▌
Brainstorm Ideas · khurrummahmood bundle
Generate a batch of candidate ledger intakes for a topic or context file, deduplicate against the existing ledger, and bulk-write the survivors as proposed-state intakes with origin AI-suggestion. The orchestrator does the creative work (reading the topic, current ledger state, project docs, and producing N candidates with title / summary / hypothesis / subsystem_kind / markers); the helper script validates against duplicates and writes. Read .claude/docs/idea-ledger.md when authoring or debugging this skill.

0
installs
▌
Find Duplication · khurrummahmood bundle
Detect and triage Python structural/lexical duplication with the legacy scout workflow, or report conservative JavaScript-family, TypeScript/TSX, exact Go/Kotlin/PHP/Ruby/Swift/Rust/Dart function-body, and exact Java method-body clone evidence. Each language uses a separate family-local pipeline and copied-skill runtime.

0
installs
▌
Propose Boundary · khurrummahmood bundle
Turn a confirmed or suspected missing-boundary into a read-only boundary-extraction proposal. Family-local evidence supports Python, TypeScript/checked-JavaScript, Go, Java 17, Kotlin/JVM, C#, Swift 6.3.3, PHP, Ruby, bounded Rust, and bounded Dart. It emits reports/propose-boundary/<target-slug>/proposal.md with candidate seams, public API, compatibility plan, caller impact, and characterization/native-verification plan. Read-only — no edits. Hands off to /refactor-subsystem (decomposition mode).

0
installs
▌
Find Route Sprawl · khurrummahmood bundle
Detect route prefixes for a product workflow that are scattered through flat URL modules, missing include ownership boundaries, or duplicated as ambiguous API aliases. SUSPECT skill for product topology.

0
installs
▌
Harvest Learnings · khurrummahmood bundle
Distill a host project's forged learnings (lint suite, precedents, ADRs, idea ledger, known-issues) AND latent production-grade standards the project skips into PORTABLE standards — each carrying an attributed exemplar + durable host back-link, a portability verdict (ports / stays-home via a translation test), and a lifecycle×stakes activation tag (ADR 0020). Recurring; two-source (extractive + generative); proposal-only — never edits code or auto-adopts. Use to mine a project for cross-project standards, refresh the portable corpus, or convert one-off discoveries into reusable guardrails.

0
installs
▌
Project Interview · khurrummahmood bundle
Build the human-approved project profile that complements /adapt-project discovery. Reads repository facts first, then interviews the user about purpose, maturity, critical workflows, risk posture, desired direction, intentional tradeoffs, known-bad legacy patterns, and do-not-break surfaces. Writes draft artifacts under reports/project-interview/scan-<TS>/ by default; durable .engineering/project/profile.yml, profile.md, and open-questions.md require --apply.

0
installs
▌
Extract State Type · khurrummahmood bundle
Read-only EXPLAIN skill that converts a function operating on an implicit mutable dict (or ad-hoc kwargs/namespace) into a typed @dataclass or TypedDict migration plan. Produces reports/extract-state-type/<target>/proposal.md with the current-shape table, proposed type definition, caller-by-caller migration plan, characterization-test matrix, and stop condition. Hands off to /fix-workflow or manual migration.

0
installs
▌
Find Comment Drift · khurrummahmood bundle
Advisory SUSPECT scan for comments, docstrings, JSDoc, and template comments that have drifted from the code they are meant to clarify. Flags detached section banners, narration comments, missing or thin public class docstrings, stale terminology, JavaScript and TypeScript functions that deserve real JSDoc, thin ceremonial JSDoc, noisy HTML comments, fragile doc references, bounded Go, Java, Kotlin/JVM, PHP, Ruby, Swift, Rust, C, and C++ lexical-comment surfaces, and bounded Dart adjacent-doc/fixed-return syntax.

0
installs
▌
Find Standard Gaps · khurrummahmood bundle
Detect places a declared baseline standard should apply but doesn't. Family-local scanners cover Python plus narrow syntax-only JavaScript/TypeScript, Go, Java, Kotlin/JVM, PHP, Ruby, Swift, Rust, and Dart direct-call guard conditions, then report every site whose triggering situation holds but the declared standard is absent. Detection-only; never edits code.

0
installs
▌
Prevent Regression · khurrummahmood bundle
Turn a closed cleanup cluster into a permanent guardrail: diff-scoped lint when the invariant is lexical/AST-shaped, or focused regression tests when the invariant is behavioral/contract-shaped. Accepts a cluster ID from a recent scan (`cluster:<id>`, `delete:<id>`, `semantic:<id>`, `dormant:<id>`), a free-form pattern description with optional `path:<glob>` qualifier, a product-topology template (`topology:<route-ownership|workflow-registry|frontend-boot|site-endpoints|doc-route-drift>`), or `--dogfood <rule-name>` to rewire an existing rule. Emits a proposal at `reports/prevent-regression/<id>/` containing the guard artifacts, canonical-pattern/testing guidance when needed, and a verification recipe. Read-only against production code — never installs the guard unilaterally; the human reviews and executes.

0
installs
▌
Refactor Subsystem · khurrummahmood bundle
Execute a large-subsystem refactor (split a bloated module, extract a service, reorganize a domain) using an orchestrator/scout pattern driven by a spec file in `ai-docs/specs/`. Operates in seven phases — Inventory, Characterize+Extract, Plan, Approve, Execute, Verify, Crystallize — with mandatory three-output scouts (primary brief + findings.md + extracted-behaviors.md), spec-first enforcement via `scripts/specs.py`, ledger updates via `scripts/ledger.py`, and a conservative default that unknown code STAYS until explicit human approval. Refactors are behavior-preserving; bug fixes surfaced along the way become separate commits.

0
installs
▌
Find Contract Drift · khurrummahmood bundle
Advisory SUSPECT scan for cross-layer contract drift on configured product-workflow targets or explicit paths: endpoint keys, DOM IDs, `data-*` selectors, template-called JS exports, and JSON/boot payload fields. Imports the existing `/find-frontend-contract-drift` boot-global detector and adds route-surface contract bands.

0
installs
▌
Find Implicit State · khurrummahmood bundle
Detect Django stringly-typed state and tuple-inferred identity patterns, plus narrow TypeScript, checked-JavaScript, Go, Java 17, Kotlin/JVM, PHP, Ruby, Rust, Dart, and Swift state branches. Semantic branches distinguish first-party bare state operations from typed authorities and excluded or unresolved evidence. Detection-only — never edits production code.

0
installs
▌
Find Orphaned Ideas · khurrummahmood bundle
Detect ideas that need attention but are not getting it. Seven modes — stale (in-flight, no event in N days), harvest (has-more-potential, not in-flight), plan-dropouts (in a plan file but missing from ledger), todo (TODO/FIXME orphans in source files), stale-plans (non-terminal plans > N days silent without active ledger tracking), dead-prototype (orphan routes/templates from a /find-dormant report), attention-gap (importance-weighted audit per ADR 0016, reads `.engineering/docs/importance-map.md`). Read-only audit by default; can optionally write `stalled` transition events when --apply-stale is set. Read .claude/docs/idea-ledger.md when authoring or debugging this skill, `.engineering/docs/todo-tuning.md` when calibrating --todo, and `.engineering/docs/importance-map.md` (plus ADR 0016) when calibrating --attention-gap.

0
installs
▌
Find Perimeter Gaps · khurrummahmood bundle
Audit the quality perimeter — report every (code root × language) cell with significant LOC and the SUSPECT detectors covering it; flag cells no detector covers. The detector fleet scans what it scans; this skill reports the inverse, so blind spots are a visible decision instead of an accident. Detection-only; never edits code. Born from a real incident — 34.6K lines of JavaScript invisible to an ecosystem whose omnibus detector was Python-only (ADR 0032).

0
installs
▌
Find Query Mutation · khurrummahmood bundle
Detect read-named methods that mutate persisted state. Runs an AST scan for functions named `get_*` / `fetch_*` / `load_*` / `list_*` / `find_*` / `check_*` whose body calls `.save()` / `.delete()` / `.update()` / `.create()` / `.bulk_create()` / `.bulk_update()` / `.update_or_create()` / `.get_or_create()`; collapses hits per function, fans out scout sub-agents to bucket each candidate (rename / split / legitimate warming / stdlib false positive), and produces a report that hands off to `/fix-workflow cluster:<symbol>`. Detection-only — never edits production code.

0
installs
▌
Find Doc Route Drift · khurrummahmood bundle
Compare documented product routes and redirect claims against Django URL patterns and simple redirect calls. SUSPECT skill for stale docs and route contract drift.

0
installs
▌
Find Layer Violation · khurrummahmood bundle
Detect views, tasks, or other entry points that own business logic. AST-walks the target for core signals (fat body, domain loop over queryset, direct LLM/agent call, dispatch bypass via bare `.delay()`, multi-model write in one function), then asks scouts to judge remaining authority in the entry point: raw SQL/cursors, direct external API clients, import/export construction, direct model/file writes, task dispatch, and transaction/resource policy. Produces an extract-service candidates report. Never edits code — hands off to `/fix-workflow layer:<candidate_id>`.

0
installs
▌
Find Stale Artifacts · khurrummahmood bundle
Detect stale working artifacts — abandoned plans, proposed plans untouched past a soft staleness budget, aged `scan-*` report directories, and orphan top-level report files. Defends attention against the accumulation of working artifacts that outlive the work. SUSPECT skill, sibling to /find-dormant for non-code surfaces.

0
installs
▌
Map Product Workflow · khurrummahmood bundle
Map a user-visible product workflow across routes, views, templates, JavaScript, docs, status providers, and compatibility redirects. Produces `.claude/docs/workflows/<name>.md`. MAP skill for topology-level quality.

0
installs
▌
Find Incomplete Sweep · khurrummahmood bundle
Advisory SUSPECT scan for incomplete sweeps — multi-site changes that were started but never finished, leaving a forgotten sibling call site at the old shape ("updated N-1 of N"). Python retains the keyword-argument omission band; TypeScript/TSX and checked JavaScript use the host-pinned TypeScript Compiler API to group resolved project function calls by object-option property presence; Go uses host Go `go/types` for one direct top-level function / keyed struct-option shape; Java 17 uses the JDK compiler tree API for one direct record/options-constructor shape; PHP uses bounded Composer PSR-4 direct constructions; Ruby requires project-authored RBS constructor contracts; Kotlin/JVM uses pinned direct constructor-call facts; Rust uses compiler-resolved direct calls for one struct-option omission shape; Dart uses SDK-LSP-resolved top-level calls for one named-argument omission shape; Swift uses compiler-AST-resolved direct calls and default-argument ownership. Gated on a git-trajectory signal: a divergence coun

0
installs
▌
Mature Existing Ideas · khurrummahmood
Research an existing ledger entry (or a batch of entries carrying the needs-research marker), append research-log notes to the ledger through /track-idea, and optionally clear markers as evidence accumulates. The orchestrator does the research (project doc grep, optional external Web/Context7 lookups, prior-art surveys) and uses track-idea/scripts/track.py for every ledger write. Read .claude/docs/idea-ledger.md when authoring or debugging this skill.

0
installs
▌
Extract Existing Ideas · khurrummahmood bundle
Walk existing prose surfaces (BACKLOG.md, lessons.md, plan files) and propose candidate ledger intakes for the items already named there. Read-only by default — emits a candidates JSON and a report. Hands off to /brainstorm-ideas (via its helper script) for the actual write, so the dedup, validation, and origin discipline stay in one place. Read .claude/docs/idea-ledger.md when authoring or debugging this skill.

0
installs
▌
Find Concept Divergence · khurrummahmood bundle
Advisory SUSPECT scan that surfaces concept-name drift between the canonical glossary at `.claude/contracts/concepts.yaml` and the codebase / docs. Detects three drift bands: `avoid_term_hit` (code uses a phrasing the glossary explicitly says is wrong), `competing_term_coexistence` (a `flagged_ambiguities` entry has multiple `competing_terms` present in the same file — drift the glossary author has not yet resolved), and `superseded_co_occurrence` (both a deprecated name and its `superseded_by:` replacement appear in the same file — rename transition drift). Detection only — never edits code or docs.

0
installs
▌
Find Dead Route Surface · khurrummahmood bundle
Advisory SUSPECT scan for configured product route-surface reachability: routes, templates, static JS, inline handlers, prototype pages, and orphaned page/API surfaces. Reuses `/find-dormant` URL extraction where useful, but focuses on the route/template/static wiring in configured targets or explicit paths.

0
installs
▌
Find Rule Surface Drift · khurrummahmood bundle
Detect drift on the agent-rules surface — oversized always-loaded files, dormant load-on-demand docs, missing-but-referenced docs, unreferenced docs, and broken links from the human onboarding entry point. SUSPECT skill governing the placement convention defined in ADR 0005 (agent-rules-design).

0
installs
▌
Find Skill Intent Drift · khurrummahmood bundle
Advisory SUSPECT scan that compares each skill under `.claude/skills/` against its machine-checkable intent + provenance contract under `.claude/contracts/skills/<name>.yaml` (schema v2). Flags four drift bands — missing (skill has no contract), orphaned (contract for a deleted skill), malformed (contract missing required schema keys), and stale (SKILL.md committed after the contract's last commit, so captured intent may have silently drifted). This is the "no easy reversion of intent" guard: it cannot block an edit, but it makes intent erosion visible. It is the intent/provenance layer of the skill-meta trio, distinct from frontmatter-contract lint and artifact-coherence drift.

0
installs
▌
Extract Cotton Primitive · khurrummahmood bundle
Turn a confirmed frontend duplication candidate into a cotton primitive proposal. Consumes a finding from /find-frontend-duplication (or an explicit category target like `modal-shell`) and emits reports/extract-cotton-primitive/<target>/proposal.md with the `<c-vars>` declaration, primitive body, callsite migration table, JS-partner notes, and stop condition. Read-only — no template/JS edits. Hands off to /refactor-subsystem or manual extraction.

0
installs
▌
Find Complexity Hotspots · khurrummahmood bundle
Detect advisory Python, JavaScript, TypeScript, Go, Java, bounded Kotlin/JVM, PHP, Ruby, Swift, Rust, and Dart function-complexity hotspots without changing production files. Preserves the Python stdlib AST scan and adds syntax-only family-local high-branch findings for bounded named functions, methods, and constructors.

0
installs
▌
Find Workflow State Gaps · khurrummahmood bundle
Advisory SUSPECT scan for workflow state coverage gaps in JavaScript and templates: loading, empty, failure, retry/cancel, disabled, and mobile states. Also imports workflow-duplication findings as state-authority context.

0
installs
▌
Extract Workflow Registry · khurrummahmood bundle
Propose a canonical registry for a mapped product workflow: step id, label, route, visibility, status provider, active tab, template owner, frontend boot keys, and registry-owned endpoint payloads. EXPLAIN skill; no code edits.

0
installs
▌
Find Frontend Duplication · khurrummahmood bundle
Detect duplicated UX shells, hand-rolled primitives, and JS helper forks across templates/ and static/js/. Compares Tailwind class chains and JS function definitions against the existing cotton primitive inventory to surface "should be a c-primitive but isn't" candidates. Hands off to /extract-cotton-primitive for proposals.

0
installs
▌
Find Semantic Duplication · khurrummahmood bundle
Detect behavioral duplication in Python functions through a scout triage pipeline, or produce conservative TypeScript/TSX, checked-JavaScript, Go, Java, bounded Kotlin/JVM, PHP, Ruby, Rust, Dart, and Swift function-level leads using host-native semantic facts. Semantic branches report bounded static candidates with human review; they do not infer behavioral equivalence, workflows, structural duplication, or safe refactors.

0
installs
▌
Find Skill Artifact Drift · khurrummahmood bundle
SUSPECT scan for instruction-artifact coherence: a SKILL.md that promises scripts, flags, tools, or evidence its files no longer provide. Band A is deterministic reference integrity (a documented script that exists nowhere, a documented --flag the script's argparse never defines, a bash block with no Bash in allowed-tools) and is safe to gate a commit on; Band B is heuristic structural proxies (orphan scripts, declared produces/evidence_required the body never wires in, a read-only not_for contradicted by editing tools) and stays advisory.

0
installs
▌
Find Workflow Duplication · khurrummahmood bundle
Detect duplicated product-step authority for the host-declared workflow labels, tab IDs, and route literals listed in `.engineering/docs/product-workflows.md`. The deterministic scanner reads the descriptor's `## Text-file globs`, classifies active executable owners versus docs/tests/routes/legacy surfaces, and reports repeated values that can drift across backend, templates, loaded JS, and docs. Detection-only; never edits production code.

0
installs
▌
Find Async Lifecycle Drift · khurrummahmood bundle
Advisory SUSPECT scan for polling, job, cancel/resume, and export lifecycle drift on configured product-workflow targets or explicit paths: unguarded polling, missing terminal handling, stale response hazards, duplicate job starts, and missing retry/cancel controls.

0
installs
▌
Find Folder Topology Drift · khurrummahmood bundle
Read-only SUSPECT audit for Python folder-topology drift and narrow, explicit-root JavaScript-family, TypeScript, Go, Java, Kotlin/JVM, PHP, Ruby, Swift, Rust, or Dart filename clusters. Python retains its ADR 0006 promotion and demotion bands; JavaScript, TypeScript, Kotlin, Rust, Dart, PHP, Ruby, and Swift use bounded filename-prefix rules, Go uses its first `_` token, and Java uses a leading CamelCase token. Use when a source folder is hard to skim because sibling filenames visibly name the same domain.

0
installs
▌
Find Test Obligation Drift · khurrummahmood bundle
Advisory SUSPECT diff analyzer that maps touched files to expected verification tiers from `.claude/docs/testing.md` and `development-workflow.md`, then flags missing nearby test/smoke obligations.

0
installs
▌
Find Transaction Overreach · khurrummahmood bundle
Detect Django `transaction.atomic()` blocks (and `@transaction.atomic` functions) that hold a DB connection while doing slow / external work. Runs an AST scan for HTTP calls, AI/SDK calls, cloud uploads, `time.sleep`, subprocess, and Celery dispatch inside atomic regions; collapses hits per block, fans out scout sub-agents to bucket each candidate (narrow / split / defer / legitimate / false positive), and produces a report that hands off to `/fix-workflow cluster:<symbol>`. Detection-only — never edits production code.

0
installs
▌
Organize Project Structure · khurrummahmood
Arrive at an ideal or near-ideal repo folder structure and organization approach under framework/tool/human constraints. Use when a project has historical top-level folders, source/input/output dumps, KB/spec/eval/runtime boundaries, or a proposed directory map that needs recursive folder summaries, ideal-vs-constrained topology review, boundary discovery, folder-worth judgment, deterministic move-plan options, dry-run validation, and a safe implementation approach. Not for one-off file moves or Python package prefix clusters.

0
installs
▌
Check Ecosystem Consistency · khurrummahmood bundle
Diff-aware ecosystem consistency audit for engineering-skills. Snapshots skills, shape routing references, docs skill-count claims, and catalog coverage; compares with the last reviewed state so significant skill changes surface follow-up obligations such as updating /which-shape.

0
installs
▌
Find Frontend Contract Drift · khurrummahmood bundle
Detect implicit template-to-JS boot contracts: scattered `window.*` globals, undeclared JS reads, repeated reads that should be accessed through a canonical payload, and globally-loaded JS auto-init that fetches or mutates DOM without a page marker.

0
installs
▌
Propose Folder Reorganization · khurrummahmood bundle
Turn a confirmed Python, Go, Java 17, Kotlin/JVM, C#, Swift 6.3.3, TypeScript, checked-JavaScript, PHP, Ruby, bounded Rust, or bounded Dart folder-topology cluster into a per-cluster reorganization proposal. Typed-source v1 resolves import impact, records compatibility and convention constraints, and emits a read-only move/test plan. No file moves or edits; hand off only after human review.

0
installs
▌
Drifty Skill · khurrummahmood bundle
A fixture skill whose SKILL.md drifted from its artifacts.

0
installs
▌
Clean Skill · khurrummahmood bundle
A fixture skill whose SKILL.md matches its artifacts.

0
installs

79 published skills