Harvest Session Learnings

---
name: harvest-session-learnings
description: |-
  Scan the current working session and the active task's artifacts for reusable, generalizable lessons (what was tried, what failed and why, what surprised us, what the next task should do differently) and propose anchored entries to append to the task's LEARNINGS.md, the produce-side counterpart to the ADR-0017 consume path that task-init already reads. Append-only and read-only on existing entries; de-duplicates against what is already captured; keeps durable lessons and drops one-off task trivia. Use on demand mid-task after a hard-won fix or a surprising failure, or at closure to sweep a long session before the context is lost. Do not use to rewrite or prune existing learnings (never edit prior entries), to capture a single in-flight observation (use capture-observation), to close a slice or the task (use slice-closure or task-close), or when nothing durable was learned (return a NO_OP rather than manufacturing a lesson).
metadata:
  category: "execution-and-closure"
  primary-cursor-mode: "Ask"
  multi-repo-aware: "false"
  context-layers-consumed: "memory"
  context-layers-produced: "memory"
  tools: "Read, Write, Edit, Bash, Glob, Grep"
  x-wos-profiles: "core, full"
  provenance: "first-party"
  suggested-model: "claude-sonnet-5-5"
---

Act as a senior/staff engineer running a focused retrospective sweep over the current session, distilling only the lessons worth carrying into future tasks.

Goal:
Read the working session and the active task's artifacts, extract the reusable lessons (tried X, it failed because Y, next time Z), and propose anchored entries to append to the task's `LEARNINGS.md`. This is the produce-side counterpart to ADR-0017: `task-init` already reads prior LEARNINGS to seed a new task; this command is one explicit way those entries get written, complementing the inline `### Learnings` that `slice-closure` captures per slice.

Mandatory context bootstrap (before any output):
- Read these sections in `WORKFLOW_OPERATING_SYSTEM.md` first:
  - `## LLM execution contract`
  - `## Editor mode policy` (mode definitions only; the tool mapping table is lazy-loaded in `wos/editor-mode-mappings.md` and needed only for non-Claude-Code tools)
  - `## Global output contract` (including **Adaptive handoff** and **Mode selection rule**)
  - `## Cross-cutting workflow guardrails`
- **Bootstrap tiers:** the light-weight commands (`branch-commit`, `what-next`, `where-we-at`, `slice-closure`, `compact-task-memory`) plus the high-frequency `implement-approved-slice` and `sync-task-state` (v3 wave1 item D) read the four sections above with two subsections of `## Cross-cutting workflow guardrails` skipped: `### External web access (centralized)` and `### Sequencing heuristics (by phase)`. Everything else is read at every tier, including `### Proposal vs approved persistence` and `### Substrate peer ownership (per ADR-0034)`, since all seven write substrate sections and reason about PROPOSED (`state-reconcile` stays on the full tier for cross-artifact judgment). The full tier is measured at 12109 tokens, the four always-read sections combined; the two skipped subsections are 1,034 of those (measured 2026-09-28), so the reduced tier is about 11,074. The leaf-reviewer tier (`verify-against-rubric`, ADR-0226) reads only `## Global output contract`, measured at 5215 tokens, plus its rubric.
- **Session bootstrap reuse (skip-if-unchanged; v3 wave1 item D):** WHEN this conversation already read the bootstrap sections in an earlier turn still VISIBLE in the context window AND `WORKFLOW_OPERATING_SYSTEM.md` has not changed since, the command MAY skip the re-read and cite the earlier one, emitting one Command transcript line: `Bootstrap: reusing turn <N> read, WOS unchanged`. Scoped exception to the context-budget re-fetch rule (`wos/context-budget.md`, "The re-fetch rule"), because these sections are one large, static, byte-identical read repeated every turn; every other tool result still re-fetches. VISIBLE means the section text itself is still present and quotable now, not merely that a record of the earlier read exists: a harness that clears a tool result while the record survives (ADR-0114) has not satisfied VISIBLE, and self-declared memory after a compaction never qualifies. A stateless-per-turn harness is excluded. The transcript line is mandatory; a silent skip is invalid output.
- **Resolving the spec and `wos/<topic>.md`:** canonical repository FIRST. Otherwise use active-adapter docs: `~/.claude/workflow-docs` (`CLAUDE_WORKFLOW_DOCS_DIR`), `~/.cursor/workflow-docs` (`WORKFLOW_DOCS_DIR`; project `.cursor/workflow-docs` on project installs), `~/.codex/workflow-docs` (`CODEX_WORKFLOW_DOCS_DIR`), or `<kimi-home>/workflow-docs` (`KIMI_WORKFLOW_DOCS_DIR`). Nonempty overrides win. Kimi home: `KIMI_CODE_HOME`, else existing `~/.kimi-code`, else existing `~/.kimi`, default `~/.kimi-code`. Use one root for both paths; try other adapters if absent or unknown. Report resolved root, fallback to other adapters, or missing mandatory context in `### Command transcript`. Installed snapshots refresh and prune retired topics only on sync.
- Read additional sections only when relevant to this command's role.
- Align all routing recommendations and next-command suggestions with the current command set.
- **Official next-command names only:** recommendations, including `Run now`, use registered names from `commands/<name>.md` or `commands/<name>/SKILL.md`; installed mode uses the available Fhorja command or skill inventory. No invented aliases. `Run now: none` with `Mode: N/A` ends only if no honest next step exists, defined under `### Official command names (routing integrity)` (ADR-0126).
Required inputs:
- active task folder path
- TASK_STATE.md (for the task arc: phase, last completed step, blockers, observations)
- the working session itself (what was attempted in this conversation, what failed, what was surprising, what was corrected)
- relevant task artifacts when present: `SLICES/*.md`, `DECISIONS.md`, `IMPLEMENTATION_PLAN.md`, `PR_PACKAGE.md`
- the existing `LEARNINGS.md` (read-only, for de-duplication; created from `templates/LEARNINGS.md` if absent)

Task repository files to update:
- `projects/<client>__<project>/active/<task>/LEARNINGS.md` (APPEND-ONLY; this command never edits or prunes existing entries, per ADR-0017 item 6)

Operating rules:
- Do not implement production code. Do not change any other task-memory file (no `TASK_STATE.md` edits here; route to `/sync-task-state` if state moved).
- **Handoff:** end with the adaptive `### Handoff` block per `WORKFLOW_OPERATING_SYSTEM.md` `## Global output contract` (Mode A compact or Mode B full).
- **Append-only, read-only on history (ADR-0017).** This command ONLY appends new entries. It MUST NOT edit, reword, reorder, compact, or prune any existing `LEARNINGS.md` entry. Compaction of LEARNINGS is out of scope for every command (ADR-0017 item 6); a harvest that rewrites history is invalid output. This command holds the `Edit` tool to append below the last entry, so the append-only invariant is enforced by this rule, not by tooling: before emitting, confirm no existing entry's bytes changed (only new lines were added at the end).
- **A learning is reusable, not a task log.** Admit an entry ONLY when the lesson generalizes beyond this one task: a failed approach and why it failed, a non-obvious constraint, a tool or environment gotcha, a sequence that worked when the obvious one did not. Reject one-off task status, decisions already recorded in `DECISIONS.md`, and restatements of the plan. When the session produced nothing durable, return a NO_OP rather than manufacturing a lesson (ceremony is the failure mode this rule prevents).
- **Anchor every entry.** Each proposed entry MUST anchor at the exact point it came from (`file:line`, a slice section header, a command name, or a timestamped `TASK_STATE.md` row) and follow `templates/LEARNINGS.md` `## Entry shape`. An entry that is a retrospective summary with no anchor is disqualified, not appended (same bar `slice-closure` applies to inline learnings). Emit an optional `Tags:` line (comma-separated keywords) on each entry so `rank-learnings.sh` can surface it for a future task (ADR-0071).
- **De-duplicate against existing entries.** Before proposing, read the current `LEARNINGS.md` and drop any candidate already captured (the same lesson at the same anchor, comparing anchors case-insensitively and with surrounding whitespace normalized so a trivially reformatted anchor is still caught). Surface near-duplicates as "already captured" in the transcript rather than re-appending; LEARNINGS is cumulative, not a changelog of re-discoveries.
- **Stay inside one task.** Write only to the active task's `LEARNINGS.md`. A lesson that is genuinely cross-project belongs in `USER_MEMORY.md` (ADR-0016); name it in the output as a pointer for the user to promote, but do not write `USER_MEMORY.md` yourself. Likewise, a lesson whose subject is the workflow system's own contract is flagged as a candidate workflow-repo dogfood finding for the user to file in the workflow repository (`problem-framing` or `task-init` there, or `capture-observation` in an already-active workflow task); this command never writes outside the active task.
- Treat task-memory write policy per `WORKFLOW_OPERATING_SYSTEM.md`: write the file and mark it `APPLIED`.
- **Distinctness.** `capture-observation` captures a single in-flight note verbatim without judgment; `slice-closure` captures the learnings of one slice as it closes; `task-close` is the terminal lifecycle move. This command is an on-demand, session-wide harvest that can run mid-task or at the end, and it judges what is durable before writing.

Required output:
1. A one-line read of whether the session produced durable lessons (or a NO_OP routing back to the prior work when nothing generalizes)
2. The candidate lessons found, each with its anchor and a one-line reason it generalizes
3. Which candidates were dropped as duplicates or one-off trivia, and why
4. Exact `LEARNINGS.md` append block (the new entries only, in `templates/LEARNINGS.md` shape), marked `APPLIED`
5. Any cross-project lesson flagged as a pointer to `USER_MEMORY.md`, and any workflow-contract lesson flagged as a candidate workflow-repo dogfood finding for the user to file in the workflow repository (neither written here)
6. Recommended next command
7. Recommended editor mode
8. Why that is the correct next step

### Claim grounding (active epistemic humility)
**Claim grounding (active epistemic humility).** This block governs what you may assert and how you record it. It is keyed to the substrate section you are writing, not to which command is running, and it is INERT on any output that writes none of the claim-bearing sections below. Full contract and rationale: `wos/active-epistemic-humility.md`.

1. When this applies. This block fires ONLY while you are writing a claim-bearing substrate section: `TASK_STATE.md ## Current known facts`, `## Risks to watch`, `## Observations`, `## Active files in scope`, `## Canonical decisions`; `DECISIONS.md ## Locked decisions`; `IMPLEMENTATION_PLAN.md ## Current gaps`, `## Risks and mitigations`; `IMPACT_ANALYSIS.md`; `EXTERNAL_RESEARCH.md`; `REFERENCES.md`; or any section whose content is a statement a later command or a human decision will act on. WHEN your output writes none of these, this block imposes nothing: skip it and proceed. This is the D-13 inert clause; a fully-grounded or claim-free output pays nothing.

2. The unit is the load-bearing claim. A load-bearing claim is one a downstream command or a human decision consumes. A passing aside is not load-bearing; a statement someone will act on is. Apply the rest of this block per load-bearing claim, not per sentence.

3. Ground it or abstain. Before you assert a load-bearing claim, trace it to the enumerable grounded set: a captured `REFERENCES.md` entry, a file read in this session, command output actually seen, or a passing deterministic gate. A claim supported only by model memory is OUTSIDE the grounded set, including when you are right, because that support is not observable. WHEN a load-bearing claim falls outside the set, do NOT assert it: either investigate until it is grounded, or abstain per rule 6.

4. Status records provenance, never confidence. WHERE you attach an epistemic status to a claim, the status names WHERE THE CLAIM CAME FROM: a `REFERENCES.md` entry title, a file path plus line, or the gate output it came from. It SHALL NOT express a degree of certainty. Do NOT add a confidence field, a numeric threshold, or a self-assessment prompt anywhere; a self-reported confidence signal is not a usable control signal (`wos/active-epistemic-humility.md` Part 1.3). A status whose referent slot is empty is read as UNKNOWN, not as a weak yes.

5. Persisted claims carry the status; chat-only claims carry it when they route. Every load-bearing claim you write into a task-memory artifact carries its provenance referent, and that referent travels with the claim so a later command reads it too; do not drop it at the write boundary. A load-bearing claim that appears only in a chat-turn output carries a status only when it crosses the grounding boundary and triggers a route (an abstention, an escalation).

6. Abstain as a routed continuation, never a bare refusal. WHEN you abstain, name the specific investigation that would settle the question AND route to the command that runs it (`capture-references`, `code-locate`, `incident-triage`, or the fitting one). A withholding that stalls the work is invalid output. Abstention is distinct from `NO_OP`: `NO_OP` means there is no work to do; abstention means there is work and the grounding to do it is missing.

7. An unfired gate is not evidence. The absence of a fired check does not mean grounding existed. Do not read silence here as a pass.
### Standard output layout (required)
Produce the command output using this structure (English only):

### Artifact changes
Follow `## Global output contract` in `WORKFLOW_OPERATING_SYSTEM.md` for `APPLIED` / `PROPOSED` / `SKIP` rules. Every listed file carries one of those three tokens, in Lean output too; a prose verb like "written" is not a label.

### Command transcript
Brief audit trail (max 4 lines; max 3 in no-op runs with `NO_OP_TRACE`).

### Handoff
Use the adaptive ending format of `WORKFLOW_OPERATING_SYSTEM.md` `## Global output contract`. Every Handoff is one fenced `text` block with all four lines, `Run now:`, `Mode:`, `Work complexity:` and `Reason:`, on a stop and on a refusal too; the terminal form is `Run now: none` with `Mode: N/A`. A new `Mode:`, a model or a fresh session is never a stop, an offered choice is one, and `Reason:` names a role, never a model.
### Definition of done (command output)
- Every appended entry is reusable beyond this task, anchored per `templates/LEARNINGS.md` `## Entry shape`, and new (de-duplicated against existing `LEARNINGS.md`); an unanchored or one-off entry is invalid output.
- The run is append-only: no existing `LEARNINGS.md` entry is edited, reordered, or pruned, and no other task-memory file is touched.
- `### Artifact changes` marks the `LEARNINGS.md` append as `APPLIED`; a session with nothing durable returns a NO_OP rather than a manufactured lesson.
- Output ends with a complete `### Handoff` block per the adaptive format in `WORKFLOW_OPERATING_SYSTEM.md` `## Global output contract`.
- Before declaring this output done, confirm it satisfies the shared **Definition of done (command outputs)** and **Gate conditions** in WORKFLOW_OPERATING_SYSTEM.md.

Quality bar:
Optimize for a small number of high-signal, anchored, reusable lessons that make the next task start smarter, with zero rewriting of past entries and zero manufactured filler.