pedropacheco95
UserA codebase-understanding system for Claude Code: a persistent, actively-refreshed knowledge layer that makes Claude behave like it already knows your codebase.
Categories
Indexed Skills (26)
cortex-archive-ingest
Ingest any document into Cortex's archive — client specs, contracts, regulatory/compliance documents, technical specs from stakeholders, meeting transcripts, interview recordings, call notes, RFCs/architecture docs, existing codebase documentation, user research and feedback. Use this when the user says "ingest this transcript", "ingest this RFP", "ingest this brief", "ingest this email", "ingest this design doc", "ingest this client spec", "ingest this contract", "add this to project memory", "capture this call", "remember this decision", "log these stakeholders", or drops any raw document. One skill, internal type routing by `archive/types/*.yaml` — a new document type never requires a new skill. Sibling of `specflow-ingest`, which owns requirement-shaped sources destined for the spec trees. This skill replaces the retired `cortex-ingest` (atlas-only ingestion is now the `atlas` extraction strategy inside this skill).
cortex-extract-insight
Build or refresh the Cortex insight layer — the per-file, per-concept understanding of a codebase stored under `.cortex/insight/`. Use this when the user says "extract insight", "run the extraction", "build the insight layer", "map this codebase", "run the initial extraction", "Cortex doesn't understand this codebase yet", or "refresh insight for <file/scope/concept>"; when a scheduled task invokes the extraction directly; or when an insight refresh loop (cortex-loop-insight-refresh-daily, a member of the daily bundle, or cortex-loop-insight-refresh-full, a member of the weekly-quality bundle) hands this skill a dirty-file worklist to re-extract. Runs four phases: a deterministic L1 structural pass (Core), extraction planning with a scope registry, parallel per-scope L2/L3 execution, and cross-scope L4 unification. There is no `cortex extract-insight` CLI command — this skill IS the extraction.
cortex-loop-bug-triage
Daily bug-ledger triage loop for a Cortex project. Use for the scheduled bug-triage task, or when the user says "run the bug-triage loop", "triage the open bugs", or "classify the bug ledger". Runs `cortex loop-bug-triage --collect`, classifies each worklist bug in-session with the specflow-bugs discipline, runs `cortex loop-bug-triage --report`, and summarises .cortex/pulse/reports/bug-triage.md.
cortex-loop-insight-refresh-daily
Daily insight-refresh loop for a Cortex project. Use for the scheduled insight-refresh-daily task, or when the user says "run the daily insight refresh", "refresh the insight layer", or "re-extract what changed". Runs `cortex loop-insight-refresh --daily --collect`, triages the uncertain files in-session, re-extracts L2/L3 via cortex-extract-insight in dirty-only mode, re-verifies stale references and aged edges, runs `cortex loop-insight-refresh --daily --apply`, and summarises .cortex/pulse/reports/insight-refresh.md.
cortex-loop-insight-refresh-full
Weekly ground-truth insight regeneration loop for a Cortex project. Use for the scheduled insight-refresh-full task, or when the user says "run the full insight refresh", "regenerate the insight graph", or "rebuild L4". Runs `cortex loop-insight-refresh --full --collect`, re-runs cross-scope L4 unification via cortex-extract-insight over all scopes, runs `cortex loop-insight-refresh --full --report`, and summarises .cortex/pulse/reports/insight-refresh.md.
cortex-loop-session-observe
Session-observation loop for a Cortex project — the v3 successor to v2's insight-gaps, and the primary producer of automatic, ungated project learning. Use for the scheduled session-observe task, or when the user says "run the session-observe loop", "what did we learn this session", or "capture session observations". Runs `cortex loop-session-observe --collect`, reads the unobserved sessions from the shared corpus, infers durable knowledge (project context stated in passing, user corrections, gotchas hit, non-obvious behaviour, patterns established), classifies each with a prefer-ungated tiebreaker, and enriches directly with no approval — project-wide context into insight/observations/, per-file facts into insight per-file entries — carrying claude-sessions provenance; only the minority that should bind future work becomes a rule-candidate or decision-candidate pulse proposal — then runs `cortex loop-session-observe --apply --proposals <file>` and summarises .cortex/pulse/reports/session-observe.md.
cortex-loop-test-runner
Tiered test-runner loop for a Cortex project — the only code-writing loop. Use for the scheduled test-runner task, or when the user says "run the test-runner loop", "run the test cascade", or "triage the failing tests". Runs `cortex loop-test-runner --collect`, classifies each pending failure in-session with the specflow-bugs discipline, runs `cortex loop-test-runner --fix-stage <results.json>`, and summarises .cortex/pulse/reports/test-failures.md.
cortex-pulse-distil
Weekly session-distillation loop for a Cortex project. Use for the scheduled weekly-curation bundle's distil member, or when the user says "run the distil loop", "what do I keep repeating", "mine the sessions for rule candidates", or "which workflows deserve a skill". Runs `cortex pulse-distil --collect`, performs the pattern judgment in-session — including the workflow-mining lens that proposes new skills — runs `cortex pulse-distil --propose`, and summarises .cortex/pulse/suggestions.md.
cortex-register-tasks
Register a Cortex project's five scheduled-task bundles with the Claude Desktop app, from inside a Claude Desktop session, using the app's own scheduled-tasks tools. Use when the user says "register the cortex scheduled tasks", "run cortex-register-tasks", "set up the cortex loops in the app", "activate the cortex tasks", or after `cortex init` / `cortex tasks verify` reports tasks not yet registered. Reads the authoritative plan from `cortex tasks plan --json`, diffs it against mcp__scheduled-tasks__list_scheduled_tasks, creates/updates only this project's tasks (replacing its own superseded legacy tasks), and confirms with `cortex tasks verify`.
specflow-bugs
Diagnose bugs in a Specflow-managed project by walking the spec-model diagnostic tree to find the root cause layer, classify the bug type, file it in the bug ledger (.cortex/compass/bugs/), and produce a concrete change plan for the change-router to execute. Use this skill whenever the user reports a bug, describes unexpected behavior, says something is broken or wrong, or when a test failure needs triage. PROACTIVELY use this skill when you encounter: "X doesn't work", "X returns wrong result", "X crashes", "this test is failing", "the behavior should be Y but it's Z", "why does X happen", error messages, 500s, wrong data, missing responses, or any report where implemented behavior diverges from expected behavior. Also trigger when test runs produce failures that need classification — the skill handles both human-reported bugs and automated test failure triage. If a .specflow/specs/ directory exists and the user reports something broken, this skill runs first — it diagnoses before anyone touches code.
specflow-change-router
Classify every incoming user request against the project's spec tree (developer specs in `.specflow/specs/`, business specs in `.specflow/specs-business/`, and `_overview.md` folder docs) and route it to the correct action before any work begins. This skill is the mandatory entry point for ALL user interactions in a Specflow-managed project (any project with a `.specflow/specs/` directory, with or without a parallel `.specflow/specs-business/`). PROACTIVELY use this skill whenever the user says anything in a project under spec management — bug reports, feature requests, behavior changes, stakeholder questions, "what does this group do?", outcome/journey/metric talk, or anything that might touch a developer spec, a business spec, or a folder overview. Also triggers on "add a feature", "fix this bug", "change this behavior", "what does X do", "what's the goal of X", "why do we have X", and on requests that may have drifted between business and developer layers (e.g., a dev change that invalidates the business d
specflow-deep-onboard
Deep onboarding for codebases where confidence matters. Orchestrates 3 parallel specflow-onboard-codebase passes by spawning separate subagents via the Agent tool, then reads their outputs from disk, compares structurally, investigates disagreements with focused agents, and produces a final merged spec tree. This skill runs in the main Claude Code session (which has access to the Agent tool) — it must NOT be invoked as a subagent itself. For a quick single-pass onboarding, use specflow-onboard-codebase directly instead. Trigger on: "deep onboard", "high-confidence onboarding", "run 3 passes", "multi-pass onboarding", "I need confident specs for this codebase".
specflow-develop
Implement code from specs using a recursive orchestration model. Explores the codebase, researches best practices, plans the minimum implementation, codes it, and verifies via the test cascade: atomic tests at spec level, spec tests at domain level, journey tests at slice level. Self-similar at every scope — the same skill runs at slice, domain, and spec granularity. At each level: explore, plan, check size, either execute or delegate. Parent agents verify children via tests, not code review. Codes ALL gaps (including edge cases and missing validation) and documents them for human review — never stops for gaps. Use this skill when the user says "implement this spec", "build this slice", "code this domain", "develop from specs", or any request to generate code from an existing spec tree.
specflow-ingest
Read external documents (client briefs, meeting notes, requirements updates, stakeholder emails, technical specs, product docs, or any free-form input) and reconcile them against the existing spec trees to produce a structured change manifest. Uses parallel agent plans, section-based processing with delegation for large documents, internal conflict detection, and temporal/version awareness. This skill proposes changes — it does not execute them. Approved changes are routed through the change-router for execution. Trigger on: "here are the new requirements", "update specs from this", "the client sent this", "ingest this", "reconcile this with specs", "what changed from this brief", or any document drop in a Specflow-managed project.
specflow-lint
Verify and correct the structural integrity of both spec trees (.specflow/specs/ and .specflow/specs-business/). Checks that every spec follows the correct format, naming conventions, frontmatter schema, entity reference style, folder structure, and bidirectional linking rules. Fixes violations automatically where possible, flags ambiguous issues for human review. Use this skill whenever the user says "check specs", "lint specs", "verify spec structure", "are my specs correct", "fix spec formatting", "validate the spec tree", or after any bulk spec operation (onboarding, ingest, batch editing) to ensure consistency. Also trigger proactively after specflow-spec-editor completes a batch of changes, or when the specflow-viewer reports parsing warnings.
specflow-new-project
Generate a complete two-layer spec tree (business specs for stakeholders + developer specs for implementation), folder overview docs in every directory, tooling manifest, agents, skills, rules, and build order from a project description. Use this skill whenever the user wants to start a new project from scratch, describes a product idea, says things like "new project", "build me an app", "I want to create", "start a new", "specflow new", "start from scratch", "help me spec this out", "write specs for this", or presents any product concept that needs structured planning before implementation. Even if the user just describes what they want to build without explicitly asking for specs, this skill should trigger — the goal is to plan before coding. The output contains both `.specflow/specs-business/` (stakeholder-facing outcomes and journeys) and `.specflow/specs/` (developer-facing entity references, rules, acceptance criteria), bidirectionally linked, with an `_overview.md` in every folder of both trees.
specflow-onboard-codebase
Reverse-engineer a two-layer spec tree (developer specs in .specflow/specs/ plus business specs in .specflow/specs-business/, with _overview.md in every folder of both trees) from an existing codebase. Uses bottom-up atom extraction with delegated agents for deep code investigation, relationship graphing for deterministic domain discovery, adversarial investigation before any bug classification, and a verification pass against the completed specs. Use this skill when the user says "onboard this codebase", "reverse engineer specs", "generate specs from code", "bring this under spec management", "analyze this codebase", or any request to retroactively create specifications from working code.
specflow-spec-editor
Execute spec changes against the spec tree — create new specs, edit existing specs, manage bidirectional links, update folder overviews, and run coherence checks. This is the skill that actually modifies spec files. Use this skill whenever a change plan needs to be executed against specs — whether that plan comes from the change-router, from specflow-bugs, from specflow-ingest, or from a direct human instruction. PROACTIVELY trigger when: the change-router has classified a request and the next step is editing specs; a bug diagnosis produces a change plan that requires spec modification; an ingest manifest has approved changes to apply; the user says "update the spec", "add this criterion", "create a new spec", "change the rule", "add a business spec", "link these specs", "update the overview", "wire implements", "deprecate this spec", "move this spec", or any instruction that means modifying files in .specflow/specs/ or .specflow/specs-business/. If someone says "edit" or "change" and the target is a spec fil
specflow-tests
Complete testing framework for Specflow projects. Builds the test infrastructure first, then generates tests across four layers (Atomic, Spec, Journey, Scenario) via delegated agents per domain, then hands off to a separate verification agent that runs all tests, performs adversarial quality checks, and loops until everything passes. A test that does not execute is not a test. Use this skill for: generating tests from specs, setting up test infrastructure, creating test fixtures, designing scenario tests, running test suites, or auditing coverage.
specflow-viewer
Generate a polished, self-contained HTML page (`specs.html`) for browsing a Specflow project's two-layer spec tree — both the developer specs under `.specflow/specs/` and the high-level business specs under `.specflow/specs-business/`, plus the per-folder `_overview.md` docs that explain each group, all parsed and rendered as a client-facing site with a Business⇄Developer toggle, sidebar tree (folders surface their first-sentence summary as a subtitle, click a folder to read its overview), search across spec content AND folder overviews, cross-link sections (Implements / Implemented by) between the two layers, an "Unmapped" badge on dev specs missing an `implements:` link to a business spec, acceptance-criteria cards, dependency chips, and an optional test-status overlay. Use this skill PROACTIVELY whenever the user wants to show, present, export, publish, share, or hand off Specflow specs in a browsable format — including phrasings like "generate the spec viewer", "build the spec page", "make the HTML specs"
cortex-loop-atlas-staleness
Monthly project-memory staleness review for a Cortex project. Use for the scheduled atlas-staleness task, or when the user says "run the atlas loop", "is the atlas stale", or "audit project memory". Invokes `cortex loop-atlas-staleness` and summarises .cortex/pulse/reports/atlas-review.md.
cortex-loop-onboarding-drift
Monthly scaffolding-drift review for a Cortex project. Use for the scheduled onboarding-drift task, or when the user says "run the onboarding-drift loop", "is the scaffolding current", or "check the CLAUDE.md block and indexes". Invokes `cortex loop-onboarding-drift` and summarises .cortex/pulse/reports/scaffolding-review.md.
cortex-loop-rule-decay
Weekly rule-obsolescence review for a Cortex project. Use for the scheduled rule-decay task, or when the user says "run the rule-decay loop", "which rules are stale", or "audit the compass rules". Invokes `cortex loop-rule-decay` and summarises .cortex/pulse/reports/rule-candidates.md.
cortex-loop-spec-drift
Daily spec-content drift detector for a Cortex project. Use for the scheduled spec-drift task, or when the user says "run the spec-drift loop", "have the specs drifted", or "which specs lag their code". Invokes `cortex loop-spec-drift` and summarises .cortex/pulse/reports/spec-drift.md.
cortex-pulse-hygiene
Daily deterministic hygiene sweep for a Cortex project. Use for the scheduled hygiene task, or when the user says "run the hygiene loop", "hygiene sweep", "project health check", or asks what unfinished or broken state the project has. Invokes `cortex pulse-hygiene` and summarises .cortex/pulse/reports/hygiene.md.
cortex-ingest
Retired — moved into cortex-archive-ingest. Use this when the user says "ingest this transcript", "ingest this RFP", "ingest this brief", "ingest this email", "ingest this design doc", "add this to project memory", "capture this call", "remember this decision", "log these stakeholders", or drops any raw document — trigger cortex-archive-ingest instead.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.