bibennurbani
UserA structured AI workflow for planning, coding, reviewing, testing and shipping software — LLM wiki memory, a task lifecycle, risk-tiered governance, and a local cockpit that runs your agents (Claude Code, Codex, Copilot).
Categories
Indexed Skills (36)
tcgflow-audit
Run the audit-workspace skill — cross-check agents ↔ skills ↔ codebase drift in `.tcgstackflow/`. Detects broken agent→skill references, skills in the wrong library (project-local vs global per ADR 0012), skill content stale vs codebase tech versions, tech in codebase without a skill, SKILL.md frontmatter issues, tool adapter mismatches, and (for multi-project workspaces) sub-projects missing from config.yaml. Produces a report; fixes route through normal channels.
tcgflow-code
Adopt the Coder role and execute a PLANNED task. Use when the user types `/tcgflow-code [TASK-ID]` or says "implement ES-1234", "start coding the planned task", "work on X". Reads the TASK details, executes subtasks one at a time, appends YAML entries to the implementation log, surfaces HIGH/CRITICAL actions as permission requests.
tcgflow-ingest
Adopt the Ingester role and fold a Raw source into the LLM-wiki. Use when the user types `/tcgflow-ingest [TASK-ID or raw or "MCP-source description"]` or says "ingest ES-1234", "fold this into the wiki", "ingest the files in raw/". Log-first procedure — drafts wiki/log.md entry before any page changes — with approval gates for new pages and deletions. A single ingest may touch 10–15 pages.
tcgflow-init
Initialize Creative GeekStack Flow workspace in the current directory. Use when the user types `/tcgflow-init` or asks to set up geekstackflow, .tcgstackflow, or the AI workflow in this project. Runs the init.js installer with auto-detection of sub-projects (multi-project support is automatic when 2+ codebases are found at top level).
tcgflow-lint
Run the lint-wiki skill — periodic health-check of the LLM-wiki. Detects contradictions across pages, stale claims, orphan pages, important concepts missing their own page, missing cross-references, and broken wikilinks. Produces a report appended to wiki/log.md; proposed fixes route through /tcgflow-ingest. Run on demand or weekly.
tcgflow-migrate
Migrate an existing project's ad-hoc AI infrastructure (`.taskRef/`, `ai-mem/`, hand-written `CLAUDE.md`/`AGENTS.md`, scattered Copilot instructions) onto canonical Creative GeekStack Flow `.tcgstackflow/` layout. Use when the user types `/tcgflow-migrate` or wants a clean cutover from prior AI workflow scaffolding. Executes the four-phase migration pattern (init+adapters → tasks → wiki → decommission) with `.bak` backups and a CRITICAL deletion gate at the end.
tcgflow-new-feature
Adopt the Planner role and turn an idea with no ticket behind it into a PLANNED task — framing, then decomposition, in one invocation. Use when the user types `/tcgflow-new-feature {feature description}` or says "new feature", "start a new feature", "I want to build X", "frame this idea". Dispatches `frame-feature` → `grill-task` → `plan-task`, with a HIGH Jira hand-off gate before the first file is written. Interactive only.
tcgflow-plan
Adopt the Planner role and write a TASK details file for a new task. Use when the user types `/tcgflow-plan [TASK-ID-or-description]` or says "plan a task for X", "design ES-1234", "let's break down this work". Grills the user on ambiguous areas, writes the two-file task structure, updates tasks/README.md. Does not write code.
tcgflow-refactor
Adopt the Refactorer role and perform a broad, behavior-preserving refactor of a target area. Use when the user types `/tcgflow-refactor [path or area]` or says "refactor the monitoring-program module", "clean up the auth service", "do a best-practice refactor of X". Surveys the target read-only, proposes a refactor task for approval, writes characterization tests where coverage is thin, then executes behavior-preservingly and hands to the Reviewer → Tester → Ingester. Does NOT change behavior or add features.
tcgflow-review
Adopt the Reviewer role and walk an IN_REVIEW task's diff against acceptance criteria and governance. Use when the user types `/tcgflow-review [TASK-ID]` or says "review the diff", "is ES-1234 ready?", "check this work". Produces a verdict (approved / needs-work), appends a REVIEW entry to the implementation log, updates status. Never edits code — proposes changes back to the Coder.
tcgflow-session-report
Author a rich session-telemetry post-mortem (HTML) for a task from its orchestrated runs' Claude Code session logs. Use when the user types `/tcgflow-session-report [TASK-ID]` or says "write a session report for ES-1234", "where did the tokens go on X", "post-mortem the run". Reads the session JSONL(s), not just the run-record totals.
tcgflow-task-from-datadog
Generate a task from a Datadog incident, monitor alert, anomalous trace, or log spike. Four standard subtasks — investigate / mitigate / fix root cause / postmortem + monitoring improvement. Use when the user types `/tcgflow-task-from-datadog` or asks "create a task from the latest incident", "investigate the spike", "process the Datadog alert about X". Status defaults to PLANNED for retrospectives and IN_PROGRESS for live SEV1/SEV2.
tcgflow-test
Adopt the Tester role and dynamically verify an IN_TEST task — build a test plan from acceptance criteria, document it (or push to Jira), run unit/E2E/app verification, and record a pass/fail verdict. Use when the user types `/tcgflow-test [TASK-ID]` or says "test ES-1234", "verify this works", "run the E2E", "write a test plan for X". Pass → VALIDATED (Ingester); fail → IN_PROGRESS (Coder).
tcgflow-timesheet-generate
Generate this week's Tempo/Jira timesheet draft from task data plus inline admin-meeting input. Use when the user types `/tcgflow-timesheet-generate` or says "generate this week's timesheet", "draft Friday timesheet", "make worklogs for Tempo". LOW risk — only writes a Markdown file to .tcgstackflow/tasks/weekly/; does NOT submit. For submission, use /tcgflow-timesheet-submit.
tcgflow-timesheet-submit
Submit the worklogs from a generated timesheet to Jira/Tempo via the Atlassian MCP. Use when the user types `/tcgflow-timesheet-submit` or says "submit the timesheet", "push worklogs to Tempo". HIGH risk — issues a permission request per governance.md when submission_mode is approval; submits directly when submission_mode is trust. Appends a confirmation table to the timesheet file with Worklog IDs.
tcgflow-upgrade
Run `init.js --upgrade` to migrate an existing pre-v0.2 .tcgstackflow/ workspace to the current layout. Use when the user types `/tcgflow-upgrade` or asks "upgrade this workspace", "migrate to no-dotfiles convention", "rename .weekly to weekly". Non-destructive — runs layout + `wiki_search` migrations, refreshes tool-owned commands/agents (backing up drift to `.bak`), additively adds new skills, and prints a **drift report** of the existing skills + tool adapters that differ from the new templates (review/merge those, or re-check with `geekstackflow drift`). Never overwrites your tasks, wiki, governance.md, config.yaml, existing skills, or tool adapters.
tcgflow-web-test
Drive a real browser (Claude in Chrome) to verify the UI behaviour of a task or of the current uncommitted change — reads the task, its Jira ticket and the wiki, derives the surfaces to visit from the acceptance criteria plus the diff, tests the happy path and a required set of edge cases, then writes `{ID} web-test-summary.md` into the task folder with the complete executed test plan and a reproducible report per bug. Use when the user types `/tcgflow-web-test [TASK-ID] [environment/URL/notes]` or says "web test ES-1234", "check this in the browser", "click through it on UAT", "smoke test what I just changed". Interactive sessions only. Mutating a shared environment is HIGH; production is CRITICAL.
frame-feature
Interview the user about a feature that has no ticket behind it — read the wiki via qmd before asking anything, settle the problem and who has it and why now, check the user's wording against `wiki/domain.md`, produce a mandatory out-of-scope list and one checkable success measure, and record the decisions the framing settled (optionally as approved, intent-marked writes to `wiki/domain.md`, `wiki/architecture.md` and `wiki/adr/`) — until there is a brief whose acceptance criteria `grill-task` can drill into. Use when the user says "new feature", "start a new feature", "frame this idea", "I want to build X", "what should we build". Interactive sessions only — every step is a question a human has to answer.
generate-timesheet
Generate a weekly Tempo/Jira worklog draft from task files and inline admin-meeting input. Reads `tasks/active/` and `tasks/completed/` for the week, applies sugar-coating to development descriptions (always on — generic descriptions are rejected), inserts admin meetings at user-specified times, writes the draft to `tasks/weekly/Weekly_Timesheet_{YYYY-MM-DD}.md`. Does NOT submit — that's `submit-timesheet`'s job. LOW risk; runs without approval.
grill-task
Interview the user about an ambiguous task or idea until every subtask has clear acceptance criteria. Use whenever the Planner is about to write `TASK details {ID}.md` and any subtask's success condition is uncertain. Asks one question at a time, waits for the answer, and proposes a recommended answer with each question so the user can confirm-and-move-on instead of starting from scratch.
ingest
Fold a Raw source (a `VALIDATED` task, files dropped into `raw/`, or an MCP-driven investigation) into the LLM-wiki. Log-first procedure — drafts the `wiki/log.md` entry before touching any page — with a gated approval step for new pages and deletions. A single ingest may touch 10–15 pages when the change is broad. Used by the Ingester agent.
lint-wiki
Periodic health-check of the LLM-wiki. Detects contradictions across pages, stale claims newer Raw has superseded, orphan pages with no inbound links, important concepts mentioned without their own page, missing cross-references, and data gaps — plus qmd-friendliness best-practice checks (missing summary, incomplete or off-taxonomy frontmatter, poor chunking structure, missing aliases) against the ingest skill's authoring standard. Produces a report appended to `wiki/log.md` and proposes fixes — never silently rewrites. Run on demand or weekly. Used by the Ingester agent.
migrate-to-gsf
Migrate an existing project from ad-hoc AI infrastructure (a `.taskRef/`-style tasks folder, an `ai-mem/`-style memory folder, hand-written `CLAUDE.md`/`AGENTS.md`, scattered Copilot instructions) onto the canonical Creative GeekStack Flow `.tcgstackflow/` layout. Clean cutover with `.bak` backups. Four phases — Init+adapters / Tasks migration / Wiki ingestion / Decommission — plus a cross-cutting move of tech skills from project-local to global. The CRITICAL deletion step requires explicit recorded approval.
plan-task
Generate the two-file structure (`TASK {ID}.md` + `TASK details {ID}.md`) for a new task and fill the details file with a flat subtask list plus acceptance criteria. Use this after `grill-task` has resolved ambiguities. The skill creates the task folder under `tasks/active/{ID}/`, scaffolds the implementation log, writes the details file, and appends a row to `tasks/README.md`. Two-file rule is enforced strictly — never produces per-subtask files.
review-diff
Walk an `IN_REVIEW` task's diff against its acceptance criteria, the project's `governance.md`, and standard code-quality heuristics. Produce a verdict (`approved` | `needs-work`) plus a findings list, append a REVIEW entry to `TASK {ID}.md`, and update the top-level status. Used by the Reviewer agent. Never edits code — proposes changes back to the Coder.
sync-jira
Fetch the current Jira status of each Jira-keyed task in the project via the Atlassian MCP and write them to the project-local cache `tasks/jira-cache.json`, which the Cockpit reads to show each task's Jira status alongside its workspace status (and flag drift). LOW risk — read-only on Jira, writes only the local cache file. The Cockpit server has no Jira credentials; this AI-run skill is the only thing that talks to Jira.
task-from-cypress
Generate a PLANNED task from Cypress test results — failing specs, flaky tests, and pending tests. Groups by spec file, creates one task with subtasks per failure pattern. Use when the user types `/tcgflow-task-from-cypress` or asks "create tasks from failing tests", "what's flaky?", "process the Cypress run". Classifies each failure as flaky / genuinely broken / test-itself-wrong and proposes appropriate acceptance criteria.
task-from-datadog
Generate a PLANNED task from a Datadog incident, monitor alert, or anomalous trace. One task per incident/alert with subtasks for investigation, root cause, fix, and (if appropriate) postmortem + monitoring improvement. Use when the user types `/tcgflow-task-from-datadog` or asks "create a task from the latest incident", "process the Datadog alert", "investigate the spike". HIGH or CRITICAL risk by default — production telemetry rarely surfaces low-stakes work.
task-from-snyk
Generate a PLANNED task from Snyk MCP findings — security vulnerabilities and dependency issues. Groups findings by vulnerable package, creates one task with subtasks per finding. Use when the user types `/tcgflow-task-from-snyk` or asks "create tasks from Snyk", "process the latest vulnerabilities". CRITICAL/HIGH severity findings get surfaced as HIGH/CRITICAL Risk; MEDIUM/LOW are listed but not auto-tasked unless the user opts in.
update-task-log
Append a YAML entry to `TASK {ID}.md` after each meaningful change the Coder makes during implementation, and — the one non-Coder use — when the Planner backfills the discovery Jira gate's HIGH approval into a freshly written task (ADR 0045). Captures timestamp, author (claude/codex/copilot/human), summary, files touched, why, validation steps, and optional tags. The structured shape lets the Reviewer parse what happened and lets `generate-timesheet` derive sugar-coated descriptions later.
verify
Build a test plan from a task's acceptance criteria, document it (or push it to Jira), run the verification (unit + E2E + app behavior), and record a pass/fail verdict in the task log. Used by the Tester agent on IN_TEST tasks. Does not edit production code — proposes fixes back to the Coder. Pushing the test plan to Jira is a HIGH action (approval-gated).
web-test
Verify UI behaviour in a real browser — derive the scope from a task's acceptance criteria and/or the uncommitted diff, resolve the target environment and its auth state, drive the page through a browser-automation MCP (Claude in Chrome), exercise the happy path plus a required set of edge cases, capture console/network/visual evidence, write `{ID} web-test-summary.md` (executed test plan + reproducible bug reports) into the task folder, and record a WEBTEST verdict in the task log. Used by the Tester (and standalone on work-in-progress). Interactive sessions only — a headless orchestrated run must not drive a browser (each call stalls it on a governance approval card). Mutating a shared environment is HIGH; production is CRITICAL.
wiki-search
Find relevant LLM-wiki and docs/ content via qmd (local hybrid keyword+vector+rerank search) before reading or editing the wiki. The canonical discovery layer every agent uses — qmd surfaces which pages are relevant, then you open them and follow [[wikilinks]] one hop. Use when searching notes, finding related content, or retrieving documents from the indexed wiki/docs collections. Falls back to index.md Map-of-Content navigation when qmd is unavailable.
tcgflow-sync-jira
Fetch the current Jira status of the project's Jira-keyed tasks via the Atlassian MCP and write them to the project-local cache tasks/jira-cache.json, which the Cockpit reads to show Jira status alongside workspace status (and flag drift). Use when the user types `/tcgflow-sync-jira` or says "sync Jira", "refresh Jira status", "check the Jira status of our tasks". LOW risk — read-only on Jira; writes only the local cache.
tcgflow-task-from-cypress
Generate a PLANNED task from Cypress test failures and flaky specs — grouped by spec file, classified as flaky / genuinely-broken / test-itself-wrong / pending. Use when the user types `/tcgflow-task-from-cypress` or asks "create tasks from failing tests", "what's flaky?", "process the Cypress run". Dispatches to the `task-from-cypress` workspace skill.
tcgflow-task-from-snyk
Generate a PLANNED task from Snyk MCP findings — security vulnerabilities grouped by vulnerable package. Use when the user types `/tcgflow-task-from-snyk` or asks "create tasks from Snyk", "process the latest vulnerabilities", "what security issues do we have?". Dispatches to the `task-from-snyk` workspace skill which dedups against existing tasks, groups by package, and creates one PLANNED task per package with subtasks per finding.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.